WO2001093530A2 - Global load balancing across mirrored data centers - Google Patents

Global load balancing across mirrored data centers Download PDF

Info

Publication number
WO2001093530A2
WO2001093530A2 PCT/US2001/017176 US0117176W WO0193530A2 WO 2001093530 A2 WO2001093530 A2 WO 2001093530A2 US 0117176 W US0117176 W US 0117176W WO 0193530 A2 WO0193530 A2 WO 0193530A2
Authority
WO
WIPO (PCT)
Prior art keywords
given
data
map
network
name server
Prior art date
Application number
PCT/US2001/017176
Other languages
French (fr)
Other versions
WO2001093530A3 (en
Inventor
F. Thomson Leighton
Daniel M. Lewin
Ravi Sundaram
Rizwan S. Dhanidina
Robert Kleinberg
Matthew Levine
Andrian Soviani
Bruce Maggs
Hariharan Shankar Rahul
Srikanth Thirumalai
Jay Gunvantrai Parikh
Yoav O. Yerushalmi
Madhukar R. Korupolu
Original Assignee
Akamai Technologies, Inc.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Akamai Technologies, Inc. filed Critical Akamai Technologies, Inc.
Priority to CA2450394A priority Critical patent/CA2450394C/en
Priority to JP2001588195A priority patent/JP4690628B2/en
Priority to AU2001265051A priority patent/AU2001265051A1/en
Priority to EP01939545A priority patent/EP1290853A2/en
Publication of WO2001093530A2 publication Critical patent/WO2001093530A2/en
Publication of WO2001093530A3 publication Critical patent/WO2001093530A3/en

Links

Classifications

    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L69/00Network arrangements, protocols or services independent of the application payload and not provided for in the other groups of this subclass
    • H04L69/30Definitions, standards or architectural aspects of layered protocol stacks
    • H04L69/32Architecture of open systems interconnection [OSI] 7-layer type protocol stacks, e.g. the interfaces between the data link level and the physical level
    • H04L69/322Intralayer communication protocols among peer entities or protocol data unit [PDU] definitions
    • H04L69/329Intralayer communication protocols among peer entities or protocol data unit [PDU] definitions in the application layer [OSI layer 7]
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00Arrangements for monitoring or testing data switching networks
    • H04L43/50Testing arrangements
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L61/00Network arrangements, protocols or services for addressing or naming
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L61/00Network arrangements, protocols or services for addressing or naming
    • H04L61/45Network directories; Name-to-address mapping
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L61/00Network arrangements, protocols or services for addressing or naming
    • H04L61/45Network directories; Name-to-address mapping
    • H04L61/4505Network directories; Name-to-address mapping using standardised directories; using standardised directory access protocols
    • H04L61/4511Network directories; Name-to-address mapping using standardised directories; using standardised directory access protocols using domain name system [DNS]
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00Network arrangements or protocols for supporting network services or applications
    • H04L67/01Protocols
    • H04L67/10Protocols in which an application is distributed across nodes in the network
    • H04L67/1001Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00Network arrangements or protocols for supporting network services or applications
    • H04L67/01Protocols
    • H04L67/10Protocols in which an application is distributed across nodes in the network
    • H04L67/1001Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
    • H04L67/1004Server selection for load balancing
    • H04L67/1008Server selection for load balancing based on parameters of servers, e.g. available memory or workload
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00Network arrangements or protocols for supporting network services or applications
    • H04L67/01Protocols
    • H04L67/10Protocols in which an application is distributed across nodes in the network
    • H04L67/1001Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
    • H04L67/1004Server selection for load balancing
    • H04L67/101Server selection for load balancing based on network conditions
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00Network arrangements or protocols for supporting network services or applications
    • H04L67/01Protocols
    • H04L67/10Protocols in which an application is distributed across nodes in the network
    • H04L67/1001Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
    • H04L67/1034Reaction to server failures by a load balancer
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00Network arrangements or protocols for supporting network services or applications
    • H04L67/01Protocols
    • H04L67/10Protocols in which an application is distributed across nodes in the network
    • H04L67/1001Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
    • H04L67/1038Load balancing arrangements to avoid a single path through a load balancer
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04LTRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00Network arrangements or protocols for supporting network services or applications
    • H04L67/50Network services
    • H04L67/56Provisioning of proxy services
    • H04L67/563Data redirection of data network streams

Definitions

  • the present invention relates generally to high-performance, fault-tolerant content delivery and, in particular, to systems and methods for balancing loads from mirrored data centers within a global computer network.
  • the invention is an intelligent traffic redirection system that does global load balancing. It can be used in any situation where an end-user requires access to a replicated resource.
  • the method directs end-users to the appropriate replica so that the route to the replica is good from a network standpoint and the replica is not overloaded.
  • DNS Domain Name Service
  • the technique preferably uses a Domain Name Service (DNS) to provide IP addresses for the appropriate replica.
  • DNS Domain Name Service
  • the most common use is to direct traffic to a mirrored web site.
  • Other uses are to direct caches to storage servers, to direct streaming servers to signal acquisition points, to direct logging processes to log archiving servers, to direct mail processes to mail servers, and the like.
  • the method relies on a network map that is generated continuously for the user-base of the entire Internet.
  • a "data center” is typically located at a telecommunications facility that leases space and sells connectivity to the Internet. Multiple content providers may host their web sites at a given data center. Instead of probing each local name server (or other host) that is connectable to the mirrored data centers, the network map identifies connectivity with respect to a much smaller set of proxy points, called "core" (or "common”) points. A core point then becomes representative of a set of local name servers (or other hosts) that, from a data center's perspective, share the point.
  • Each set of mirrored data centers has an associated map that identifies a set of core points.
  • a core point is discovered as follows. An incremental trace route is executed from each of the set of mirrored data centers to a local name server that may be used by client to resolve a request for a replica stored at the data centers. An intersection of the trace routes at a common routing point is then identified.
  • the common routing point may be the first common point for the trace routes when viewed from the perspective of the data centers (or the last common point for the trace routes when viewed from the perspective of the local name server).
  • the common routing point is then identified as the core point for the local name server.
  • a core point is identified for other local name servers (or other hosts) in the same manner.
  • a given set of mirrored data centers may have associated therewith a set of core points that are then useful in estimating the relative connectivity to the set of data centers, as is described below.
  • ICMP or so-called "ping" packets
  • RTT roundtrip time
  • a core point may be pinged periodically (e.g., every 30 seconds) and the associated latency and packet loss data collected.
  • an average latency is calculated, preferably using an exponentially time-weighted average of all previous measurements and the new measurement.
  • a similar function is used to calculate average packet loss.
  • a score is generated for each path between one of the data centers and the core point.
  • the score may be generated by modifying an average latency, e.g., with a given penalty factor, that weights the average latency in a unique way to provide a download prediction. Whichever data center has the best score (representing the best-performing network connectivity for that time slice) is then associated with the core point.
  • a full network map is created by generalizing a core point/data center data set to an
  • IP block/data center data set This "unification" fills in and reduces the size of the network map and enables traffic redirection to be carried out for new local name servers.
  • the generated network map is then used to effect traffic redirection and load balancing.
  • the method preferably uses the network map to return to the local name server a list of web server IP addresses at the optimal data center. If ping data is not available for the user's local name server (of it the IP block has not been extended through unification), BGP or geo-routing can be used to make a default routing decision. Content provider-specified load balancing preferences may also be readily enforced across the data centers and/or within a particular data center.
  • Figure 1 is an illustration of a mirrored Web site that is managed by a global traffic manager according to the present invention
  • FIG. 2 is a high level illustration of the components of the GTM service
  • Figure 3 is a simplified illustration of a core point discovery process of the invention
  • Figure 4 is a simplified illustration of how an end user request is processed by the global traffic redirection system of the present invention for a mirrored web site that has been integrated into the managed service;
  • FIG 5 is a flowchart describing how a map is generated by the GTM system
  • Figure 6 is a simplified block diagram of one implementation of the global traffic management system of the invention.
  • Figure 7 is a representative traceroute generated during the core point discovery process.
  • GTM global traffic management
  • the global traffic management solution is a managed service provided by a service provider, such as a content delivery network
  • CDN CDN service provider
  • CDNSP CDN service provider
  • a CDN is a network of geographically distributed content delivery nodes that are arranged for efficient delivery of digital content (e.g., Web content, streaming media and applications) on behalf of third party content providers.
  • a request from a requesting end user for given content is directed to a "best" replica, where "best” usually means that the item is served to the client quickly compared to the time it would take to fetch it from the content provider origin server.
  • a CDN is implemented as a combination of a content delivery infrastructure, a request-routing mechanism, and a distribution infrastructure.
  • the content delivery infrastructure usually comprises a set of "surrogate" origin servers that are located at strategic locations (e.g., Internet Points of Presence, network access points, and the like) for delivering copies of content to requesting end users.
  • the request-routing mechanism allocates servers in the content delivery infrastructure to requesting clients in a way that, for web content delivery, minimizes a given client's response time and, for streaming media delivery, provides for the highest quality.
  • the distribution infrastructure consists of on-demand or push-based mechanisms that move content from the origin server to the surrogates.
  • An effective CDN serves frequently-accessed content from a surrogate that is optimal for a given requesting client.
  • a single service provider operates the request-routers, the surrogates, and the content distributors.
  • that service provider establishes business relationships with content publishers and acts on behalf of their origin server sites to provide a distributed delivery system.
  • a well-known commercial CDN that provides web content and media streaming is provided by Akamai Technologies, hie. of Cambridge, Massachusetts.
  • the present invention implements a managed service for global load balancing of a content provider's mirrored Web sites.
  • Figure 1 illustrates the basic implementation environment.
  • the global traffic management service 100 provides global traffic management for a content provider running a pair of mirror Web sites 102 and 104 (identified by the same domain, e.g., www.akamai.com).
  • the GTM service 100 provides improved responsiveness for end users 106 and 108 accessing the Web site by directing them to the best performing mirrored site.
  • Figure 2 illustrates the high level technical architecture of the GTM service which, as noted above, is implemented by a CDNSP or other entity (the "managed service provider") as a managed service on behalf of content providers running mirrored Web sites.
  • a preferred GTM service 200 comprises a number of components: a set of network agents 202, a set of web server agents 204, a set of map generation servers 206, and a set of name servers 208.
  • Each such component typically is a server, such as a Pentium-based box running the Linux operating system and having application software for carrying out the functions described below, or one or more processes executing on such a machine.
  • data is collected by the network agents and the web server agents and delivered to the map generation servers.
  • the map generation servers analyze the data, and at least one map server produces a map that assigns name server IP address/blocks to regions. At least one map is then uploaded to the name servers.
  • an end user requests access to a mirrored site domain being managed by the service, one of the name servers hands back an IP delegation that represents a "best" data center to which the user should connect.
  • a content provider has network agents located in or near their network segment within each respective data center that hosts the mirrored Web site.
  • a pair of network agents 202a and 202b are dedicated to the content provider in data center 203a
  • a pair of network agents 202c and 202d are dedicated to the content provider in data center 203b, although this is not required.
  • These network agents preferably share the same network connection as the content provider's web servers.
  • network agents 202a and 202b in data center 203a share network connections with the content provider's web servers 207a-c.
  • the set of network agents may be deployed in data centers in which the CDN is deployed.
  • multiple content providers may host their web sites at a given data center and share network agents.
  • a given network agent may collect data once for a first content provider at a given location and then share the data across all other content providers co-located in the same data center.
  • a data center typically is located at a telecommunications facility (e.g., Exodus, Frontier Global, UUUNet, and the like) that leases space and sells connectivity to the Internet.
  • a network agent has two (2) primary functions: running "core point" discovery
  • a core point typically is representative of a set of local name servers (or other hosts) that, from the perspective of a given network location (e.g., a data center), share the point.
  • a core point is a router on the Internet, although this is not a requirement.
  • the information collected from the core point discovery process is fed to the map generation servers on a relatively frequent basis, e.g., one every thirty (30) seconds, to identify down routes, congestion, route changes, and other network traffic conditions that may impair or effect connectivity to a data center at which a particular mirrored site is hosted.
  • a core (or "common") point is discovered as follows. An incremental trace route is executed from each of the set of mirrored data centers to a local name server that may be used by client to resolve a request for a replica stored at the data centers. An intersection of the trace routes at a common routing point is then identified.
  • the common routing point may be the first common point for the trace routes when viewed from the perspective of the data centers (or the last common point for the trace routes when viewed from the perspective of the local name server).
  • the common routing point is then identified as the core point for the local name server.
  • a core point is identified for other local name servers (or other hosts) in the same manner.
  • a given set of mirrored data centers may have associated therewith a set of core points that are then useful in estimating the relative connectivity to the set of data centers, as is described below.
  • FIG. 3 is a simplified diagram of the core point discovery process, in accordance with one embodiment of the invention.
  • the data center 300 corresponds to a data center located on the West Coast and the data center 302 corresponds to a data center located on the East Coast.
  • Data center locations are merely representative.
  • Each data center can host a mirror site for a given content provider.
  • a core point 305 is discovered as follows. An incremental trace route is executed from each of a set of mirrored data centers 300, 302 to local name servers 304, 306, 308 that may be used by a client machine 310.
  • the network agent (not shown) has executed a first set of traceroutes, between the data center 300 and the local name servers 304, 306 and 308, and a second set of traceroutes between the data center 302 and the local name servers 304, 306 and 308.
  • the network path between the respective data center and the local name server(s) contain router hops, as is well known.
  • the network agent identifies a location at or close to the intersection of the trace routes at a common routing point, which is shown in Figure 3 as a core point 305.
  • the common routing point may be the first common point for the trace routes when viewed from the perspective of the data centers 300 and 302 (or the last common point for the traceroutes when viewed from the perspective of the local name server 304).
  • the common routing point is then identified as the core point 305 for the local name server.
  • Figure 7 illustrates a representative core point discovery process trace.
  • the common routing point can lie somewhere along that common portion of the route.
  • the core point is the first common point for the trace routes when viewed from the perspective of the data centers, which is the same as the last common point for the trace routes when viewed from the perspective of the local name server.
  • the core point 305 need not be situated at the "exact" intersection of the trace routes. It can, for example, be located near or substantially near the intersection. It can also be located adjacent to the intersection, or it can be located at any nearby point such that measurements made to the point are representative of the measurements made at the intersection.
  • the network agent identifies other core points for other local name servers (or other hosts) in the same manner.
  • a given set of mirrored data centers may have associated therewith a set having one or more core points that are then useful in estimating the relative connectivity to the set of data centers, as is described below.
  • a network agent preferably runs core point discovery with some frequency.
  • a network agent also performs the function of periodically checking the core points assigned to one or more local name servers that already have been mapped. This process is now described.
  • Network agents preferably make measurements to core points using Internet Control Messaging Protocol (ICMP) (or so-called “ping" packets) to evaluate such information as round trip times (RTTs), packet loss, and number of router hops.
  • ICMP Internet Control Messaging Protocol
  • ping packets
  • RTTs round trip times
  • packet loss packet loss
  • number of router hops number of router hops.
  • ICMP Internet Control Messaging Protocol
  • a given network agent periodically "pings" a core point (e.g., every 30 seconds) and collects the associated latency and packet loss.
  • the network agent calculates an average latency.
  • the network agent calculates average latency using an exponentially time-weighted average of all previous measurements and the new measurement.
  • the network agent uses a similar function to calculate average packet loss. This calculation is described in more detail below.
  • the network agent uses the results to generate a "score" for each path between one of the data centers and the core point.
  • the score is generated, for example, by modifying an average latency with a given penalty factor that weights the average latency in a unique way to provide a download prediction. Whichever data center has the best score (representing the best-performing network connectivity for that time slice) is then associated with the core point.
  • the web server agents 204 do test downloads to either all the web server IP addresses or to the local load balancing devices to test for availability or "aliveness" of the mirrored sites (i.e., per data center mirror or web server).
  • a web server agent tests an object, e.g., a twenty (20) byte file available on the web server via an HTTP GET request, and check for errors and download times.
  • the measurements are taken periodically, e.g., every ten (10) seconds, although preferably a customer can change the timeout.
  • An IP address is declared "dead” if more than a given percentage of the web server agents are unable to download the test object within the timeout threshold.
  • the web server gents are preferably dispersed in co-location facilities, which are dispersed geographically and on a network basis. Moreover, one skilled in the art will recognize that the described functions of the web server agent could be performed by another component, such as the network agent, the map generation server, or some other server. Moreover, neither the web server agent nor its functions (such as testing the aliveness of a data center) are necessary for certain embodiments of the invention.
  • the map generation servers 206 receive data from the network agents and the web server agents and use this data to generate maps, which describe the mirrored site that is optimal for each IP address block.
  • a map is achieved by evaluating web server agent data, a time-weighted average of latency and packet loss, and BGP and geo information.
  • the map generation servers associate IP blocks with Internet "regions" such that a given map associates an IP block with a region number. Another data file is then used to associate region number to physical IP address, hi a representative embodiment, maps (which associate IP block to region #) are generated every few minutes and then uploaded to the name servers.
  • the name servers 208 hand out to the requesting end user the IP address(es) of the optimal data center.
  • the name server response have a time to live (TTL) of about five (5) minutes, although this value may be customer-configurable.
  • TTL time to live
  • the name servers are the same name servers used by the CDNSP to facilitate routing of end user requests to CDN content servers.
  • Figure 4 illustrates how a customer web site is integrated into the traffic redirection system of the present invention.
  • the customer has a distributed web site of at least two (2) or more mirrored sites.
  • the inventive system load balances multiple subdomains/properties provided they are in the same data centers. Integration simply requires that the customer set its authoritative name server 400 to return a CNAME to the GTM name servers 408, which, thereafter, are used to resolve DNS queries to the mirrored customer site. Recursion is also disabled at the customer's authoritative name server.
  • an end user 402 makes a request to the mirrored site using a conventional web browser or the like.
  • the end user's local name server 404 issues a request to the authoritative name server 400 (or to a root server if needed, which returns data identifying the authoritative name server).
  • the authoritative name server then returns the name of a name server 408 in the managed service.
  • the local name server then queries the name server 408 for an IP address, h response, the name server 408 responds with a set containing one or more IP addresses that are "optimal" for that given local name server and, thus, for the requesting end user.
  • the optimal set of IP addresses is generated based on network maps created by testing the performance of representative core points on the network.
  • the local name server selects an IP address from the "optimal" IP address list and returns this IP address to the requesting end user client browser. The browser then connects to that IP address to retrieve the desired content, e.g., the home page of the requested site.
  • FIG. 5 is a high level flowchart illustrating how data is processed in order to create a map.
  • the network agents Periodically (e.g., every thirty (30) seconds), the network agents ping each core point from each data center. This is step 500.
  • a time-weighted average of latency, and a time-weighted average of loss, is computed. This is step 502.
  • the weights decay exponentially in time with a time constant that is configurable.
  • the data is further processed to produce a score for each data center per core point.
  • each core point is then associated with the name servers for which the core point was a proxy.
  • a map generation process goes through all of the data and decides a set of candidate data centers for each name server. At this time, any data centers that the web server agents determine are not "alive" are discarded.
  • the map generation process extends its decisions with respect to name servers to decisions with respect to IP block.
  • a unifying algorithm is used to provide this functionality. This algorithm operates generally as follows. If all name servers in a BGP- geo block have agreeing ping decisions, then the decision of what data center is "optimal" is applied to the whole block. Conversely, if there is a disagreement, the block is broken up into the largest possible sub-blocks so that, in each sub-block, all the name servers agree.
  • the BGP-geo candidates may be used.
  • the map is produced with the candidate for each block. If there are multiple candidates, the assignments are made to get as close to the load balancing targets are possible.
  • the load balancing targets are defined, usually by the content provider, and these targets may be percentages (adding up to 100%) that breakdown the desired traffic amount by data center. This completes the map generation process.
  • step 502 involves generating a time-weighted average of latency and a time-weighted average of loss. More generally, this aspect of the invention provides a systematic methodology for predicting actual download times for various flow control protocols, e.g., TCP.
  • TCP is the most commonly used flow control protocol on the Internet today. Other protocols are built on top of UDP. Neither TCP nor UDP packets can be used to monitor the state of routes on the Internet, however.
  • ICMP packets are injected into the network (e.g., by the network agents), at preferred points in time, and then routed to a suitably chosen intermediate core point.
  • Latency is a measure of the round trip time (RTT) between the server and the core point. From maintaining a time-series of loss and latency, the system is able to predict effectively the amount of time it would take a client (that uses a name server associated with the core point) to initiate and complete a download from the server. The quality of this prediction is important for effective mapping because when a client makes a web request and there are multiple web servers from which to potentially server, it is important to be able to predict correctly which web server has the best connectivity. This is a difficult problem in general because the Internet is highly bursty and exhibits highly variable traffic conditions.
  • the following example illustrates how the time-weighted averages are computed in accordance with one embodiment of the invention.
  • a content provider Figure 3
  • the network agent "pings" the core point 305 from each data center.
  • the network agent stores the latency and the packet loss for each measurement made.
  • latency and loss parameters are merely representative of the types of signal transmission parameters that the network agent can track.
  • Other parameters that could be measured include any parameter helpful in determining the speed, quality and/or efficiency of a network path, such as parameters indicative of outages on paths, loss in signal strength, error-control data, route changes, and the like.
  • Table 1 illustrates an example of the type of data that the network agent gathers over the course of measurements made every 30 seconds between the data centers and the core point.
  • AverageLatency latx e ⁇ '' ic
  • the average latency for the data center 300 is computed as:
  • the network agent sums each weighted latency measurement (e.g., 31.88) and divides this sum by the sum of the weight factors (i.e., e "30 300 + e "60300 . . . etc.).
  • the exponentially time weighed average latency for the data center 300 is computed as:
  • the exponentially time-weighted average is 7.79, which differs from the computed average of 7.78. Although this difference does not appear significant in this example, it can be more significant for measurements averaged out over long periods of time, because more recent measurements will be given more weight than older measurements.
  • the network agent determines dynamically whether core points that were once considered optimal are still so, whether core points that had been performing well (for a given time period) are now degraded, and the like.
  • the exponentially time- weighted averaging helps also to smooth out aberrations over time in measured data and helps to indicate trends.
  • the time-weighted average latency for the East Coast data center 302 are computed in a similar manner.
  • the network agent computes a time- weighted average of loss in the same way.
  • time-weighted averages are then processed to produce a score for each data center per core point.
  • a preferred scoring function is as follows:
  • Score function average latency + ⁇ [max (100, average latency)]*(penalty factor) ⁇ , where the score is a value in milliseconds. Each of the values has a millisecond unit, except for the penalty factor, which is unit-less.
  • the value "100” is a floor or base-level value, which represents the usual round trip time required for a packet to travel between the East Coast and the West Coast. The floor is variable.
  • the term “max” refers to selecting either the value "100" or the average latency, whichever is greater. That value is then multiplied by a penalty factor, with the result then being added to the average latency to generate the score.
  • the penalty factor preferably is a function of the time-weighted average loss. Thus, in one illustrative embodiment, the penalty factor is some multiple of the time-weighted average loss.
  • the multiplier may be varied, e.g., as a function of percentage of loss, with the penalty factor contribution being higher for greater packet loss.
  • a scoring function such as described above that is based on time-weighted average latency weighted by a time- weighted average loss penalty factor affords a good approximation or "proxy" of the download time for an average size (e.g., 10Kbyte) file from the data center to an average end user.
  • an average size e.g. 10Kbyte
  • the file download time would be expected to vary as the file size is varied, but it has been found that the scoring function described above still tends to capture which data center of the mirrored set provides better performance. In other words, the absolute value of any given score is not as important as the data center-specific (e.g., East Coast vs. West Coast) values.
  • the network agent associates the core point with the local name server(s) for which the core point serves as a "proxy.”
  • Figure 6 illustrates the overall system architecture 600. As noted above, these processes typically run across multiple servers in the system. There are three logical grouping of these processes. First, the PingServer 602, PingProcessor 604, and TestPingServer 606 are running on the network agents located in the content provider's data centers. Second, the MapMaker 608, MapTester 610, and DBPusher 612 are running on another set of servers. However, these may also be run on the network agent machines if there is a shortage of servers in the network in which the global traffic management system operates.
  • MapNote Web 614 and MapNoteDNS 616 Another set of processes, called MapNote Web 614 and MapNoteDNS 616, run together on a relatively static set of machines for all customers of the system.
  • Processes 602, 604, 608, 610, 612, 614 and 616 typically run continuously.
  • An alert processor (not shown) detects if one or more machines on the network are non-functional and sends one or more corresponding alerts.
  • An archive process (not shown) is used to automatically log files and other system files periodically.
  • a file moving process (not shown) is used move data files.
  • Each server may also run a generic process monitor (not shown), which reports data to a service provider query system.
  • the global traffic management system 600 collects several pieces of data that results in a map being uploaded to the GTM name servers 615.
  • Core Point Discovery produces a list of IP addresses in a file (signifying the core points).
  • This file is sent to each PingServer 602.
  • PingServer process 602 running on each of the network agents that are deployed in a content provider's data center (not shown).
  • there is a pair of machines in each data center only one PingServer process is primary. The other one is running but only takes over if the primary goes down.
  • Each PingServer process 602 pings each of the core points approximately every 30 seconds.
  • the ping results are sent to the PingProcessors 604.
  • PingProcessors 604 preferably run on the same machines as the MapMakers 608, although this is not a requirement.
  • the PingProcessors 604 process the ping results and drop the data files off for the MapMaker 608.
  • MapMakers 608 also require data from the MapNote Web agents 614.
  • the MapNoteWeb agents 614 are the web server agents that do test downloads from the content provider's web servers. These tests are used to determine aliveness of the webservers in the data centers as has been described.
  • the MapMaker 608 looks at the ping data as well as the MapNote Web data and creates a top-level map for the top-level name servers.
  • the map is then sent to the MapTester 610 (which is usually running on the same machine).
  • the MapTester 610 uses test ping data from the TestPingServer 606 to check a given number of (e.g., a few hundred) IP addresses in the map. This is done to make sure the map is correct, however, this processing is optional Finally, if the map passes the test, it is queued for uploading to the name servers 615.
  • DBPusher 612 is one other process that preferably runs on the same machines as the MapMaker process 608. This process is solely responsible for pushing out a DB file to the top-level name servers 615. This DB file completes the lookup data for the top-level name server 615. That is, the map from the MapMaker 608 contains a mapping of IP block to a virtual region number. The DB file is the data file that actually has a mapping of the region number to physical IP addresses. DBPusher 612 monitors the MapNote Web data and, in case of a failure, pushes an updated DB file to the name servers. PingServer 602 is responsible for measuring RTT and packet loss to the list of core points. The list of core points determined as follows.
  • the service provider deploys servers in all of a content provider's data centers.
  • ping data is shared for all customers who co-locate at a particular data center, and the GTM service provider may simply pre-deploy servers at "popular" hosting facilities to save time in integrating new customers to use the system.
  • the PingServer process preferably is run on each of the network agents in a data center.
  • a leader election process (not shown) may be used to allow for the non-leader to take over if the primary fails within that same data center.
  • PingServer includes a process that is used to ping a list of IP addresses, which the PingServer receives from a system source. Also, before the list is pinged, any IP addresses that are on a restricted list are filtered out.
  • the primary inputs to the PingServer process are as follows:
  • Routers file the list of IP addresses that were discovered using Core Point Discovery.
  • PingServer The outputs of PingServer are as follows:
  • MapMaker creates the map for the top-level name servers. MapMaker takes the processed ping data from PingProcessor and the aliveness data from MapNoteWeb and constructs a map. This map contains a relationship between certain IP blocks and a region number.
  • the inputs to MapMaker are:
  • the outputs of MapMaker may include, for example:
  • MapTester is the last process before a map is uploaded to the top-level name servers. MapTester receives a candidate map from MapMaker. It then looks-up the mapping of a test IP addresses (that are pinged using TestPingServer, which is discussed more fully below). If the number of differences is below some threshold, then the map is deemed acceptable.
  • the map is then uploaded to the top-level name servers.
  • the inputs to the MapTester process are:
  • Map TestPingServer collects RTT and packet loss information for a small subset of IP addresses. This data is collected to ensure that the map produced by MapMaker is valid. MapTester, assuming the map is good, will then load the maps to the top-level name servers. The inputs to the TestPingServer process are:
  • the output of the TestPingServer process is:
  • DBPusher is responsible for processing the MapNoteWeb data and creating a DB file that is uploaded to the top-level name servers. Then, the top level name server will, after it has determined the region number for a give IP in the map, look in the corresponding DB file to find the right set of IP addresses to hand out.
  • the input to DBPusher is: • MapNote Web data
  • MapNote Web is run on a select number of servers for all customers using the traffic management system.
  • MapNoteWeb uses a list of target URLs (which, for example, could be stored in its configuration files) on which it performs test downloads. Preferably, these tests are simple HTTP GET requests and only time and errors are important. This data can be interpreted to determine whether or not a data center or web server is alive on dead. The download time is stored historically using a time-weighted average. Both the MapMaker and DBPusher use this data.
  • the input to MapNoteWeb is: • Targets to measure against (stored in configuration file)
  • MapNoteDNS is related to MapNoteWeb, except that instead of monitoring web servers it monitors name servers. Its basic function is to do a measurement at a name server for certain names and report the time. If no answer comes back, the process will attempt to ping to determine whether it is the name server or the network that caused the failure.
  • the inputs to MapNoteDNS are:
  • the intelligent traffic redirection system of the present invention has numerous advantages.
  • the system continuously pre-computes optimal maps for the user-based of the entire Internet (or, if desired, a given sub-portion thereof). It is able to do this effectively because the system reduces the scale of the problem by aggregating parts of the Internet and representing them with core points.
  • the system is also able to make different kinds of measurements depending upon the service being replicated. It combines these measurements for the core points into decisions which it then extends to the entire Internet using unification over a fallback partition of the IP address space using, e.g., BGP and geo information.
  • the system also is unique in its ability to balance load for cost minimization.
  • the system is able to pre-compute an optimal mapping for the entire Internet at all points in time. In addition to being extremely fast in its ability to react to bad network conditions, it is also extremely fine-grained in its response.
  • the system is able to detect bad server conditions quickly and is capable of interfacing with a multitude of local load balancers. By computing core points, the system makes superior measurements that mitigate the problem of intruding on firewalls and other detection mechanisms. Moreover, unlike the prior art, it can load balance load so as to minimize bandwidth costs.
  • the unification algorithm is advantageous in that it uses high-quality reliable information for a subspace and extends it to the entire space rather than falling back onto poorer quality information. This is achieved by utilizing the natural tree-like hierarchy of CIDR-based IP addressing in conjunction with the fact that Internet routers utilize the CIDR scheme for aggregating IP addresses to permit fast lookups.
  • the technique enables the redirection system to extend the benefits of high quality information from a subset of the entire space of IP addresses. This is of great importance because the Internet is experiencing exponential growth.
  • the unification algorithm affords the service provider with a means to deal intelligently with new hosts that are accessing the CDN for the first time. Current technologies do not possess a means of extending mapping decisions in this way. They either fall back to poor quality information or use a default technique, e.g., such as round robin, which essentially embodies no information whatsoever.
  • Predicting download times using ICMP probes and time-series techniques also provides numerous advantages.
  • the technique does not have any restriction on the range of file sizes and download types, and it makes intelligent use of ICMP probes of different sizes to effectively estimate packet loss.
  • the technique requires very little state for keeping track of the time-series and is able to quickly compute a new estimate using an exponentially time-weighted average of all previous measurements and the new measurement.
  • the inventive technique provides a general method for extracting a good predictor of download times based on ICMP probes.
  • the intelligent traffic redirection system is used to direct traffic to a mirrored Web site.
  • the inventive system and managed service can be used in any situation where an end-user requires access to a replicated resource.
  • the system directs end-users to the appropriate replica so that their route to the replica is good from a network standpoint and the replica is not overloaded.
  • An "end user" may be generalized to any respective client system, machine, program or process.
  • other uses of the system may include, without limitation, to direct caches to storage servers, to direct streaming servers to signal acquisition points, to direct logging processes to log archiving servers, to direct mail processes to mail servers, and the like.

Abstract

The invention is an intelligent traffic redirection system that does global load balancing. It can be used in any situation where an end-user requires access to a replicated resource. The method directs end-users to the appropriate replica so that the route to the replica is good from a network standpoint and the replica is not overloaded. The technique preferably uses a Domain Name Service (NS) to provide IP addresses for the appropriate replica. The most common use is to direct traffic to a mirrored web site.

Description

GLOBALLOAD BALANCINGACROSS MIRROREDDATA CENTERS
Related Application This application is based on and claims priority from Provisional Application Serial
No. 60/208,014, filed May 26, 2000. BACKGROUND OF THE INVENTION Technical Field
The present invention relates generally to high-performance, fault-tolerant content delivery and, in particular, to systems and methods for balancing loads from mirrored data centers within a global computer network. Description of the Related Art
It is known to store Web-based content in mirrored data centers and to load-balance such content requests to data centers based on network traffic conditions. Existing global load balancing products use several different approaches for building a map of nternet traffic conditions. One approach uses border gateway protocol (BGP) data. BGP-based routing, however, can be sub-optimal because the BGP data can be very coarse. Other approaches attempt to compute an optimal mapping in real-time and then cache the mapping information. This technique can lead to poor turnaround time during an initial "hit" and potentially stale mappings on successive requests. In addition, the quality of the measurement to the endpoint tends to be noisy. Because of the deficiencies of these mapping techniques, the resulting load balancing is less than effective.
Current load balancing devices are typically incapable of computing an optimal map for an entire computer network such as the entire Internet. Presently, the internet has hundreds of millions of hosts and routers. Estimating the connectivity time of the entire Internet to a set of mirrored data centers, such as by evaluating the network path between a server and each and every host or router, would be incredibly time-consuming and would consume far too much bandwidth. Such techniques, of course, are impractical when realtime routing decisions are required. Further, such measurements tend to be noisy and inaccurate, and they can annoy system administrators whose firewalls are contacted. Local name servers behind firewalls would not be reached and slow connectivity over the "last mile" (e.g., due to dial-up connections and the like) tend to confuse the connectivity picture. Consequently, there remains no efficient technique in the prior art for generating an optimal network connectivity map that can be used for providing intelligent traffic redirection in conjunction with load balancing across mirrored data centers located around the globe. BRIEF SUMMARY OF THE INVENTION The invention is an intelligent traffic redirection system that does global load balancing. It can be used in any situation where an end-user requires access to a replicated resource. The method directs end-users to the appropriate replica so that the route to the replica is good from a network standpoint and the replica is not overloaded. The technique preferably uses a Domain Name Service (DNS) to provide IP addresses for the appropriate replica. The most common use is to direct traffic to a mirrored web site. Other uses are to direct caches to storage servers, to direct streaming servers to signal acquisition points, to direct logging processes to log archiving servers, to direct mail processes to mail servers, and the like.
In a preferred embodiment, the method relies on a network map that is generated continuously for the user-base of the entire Internet. The problems inherent in the prior art are overcome by vastly reducing the dimensionality of the problem of estimating the relative connectivity to a set of mirrored data centers. A "data center" is typically located at a telecommunications facility that leases space and sells connectivity to the Internet. Multiple content providers may host their web sites at a given data center. Instead of probing each local name server (or other host) that is connectable to the mirrored data centers, the network map identifies connectivity with respect to a much smaller set of proxy points, called "core" (or "common") points. A core point then becomes representative of a set of local name servers (or other hosts) that, from a data center's perspective, share the point. Each set of mirrored data centers has an associated map that identifies a set of core points.
According to a preferred embodiment of the invention, a core point is discovered as follows. An incremental trace route is executed from each of the set of mirrored data centers to a local name server that may be used by client to resolve a request for a replica stored at the data centers. An intersection of the trace routes at a common routing point is then identified. Thus, for example, the common routing point may be the first common point for the trace routes when viewed from the perspective of the data centers (or the last common point for the trace routes when viewed from the perspective of the local name server). The common routing point is then identified as the core point for the local name server. A core point is identified for other local name servers (or other hosts) in the same manner. Thus, a given set of mirrored data centers may have associated therewith a set of core points that are then useful in estimating the relative connectivity to the set of data centers, as is described below.
Once core points are identified, a systematic methodology is used to estimate predicted actual download times to a given core point from each of the mirrored data centers. According to the invention, ICMP (or so-called "ping" packets) are used to measure roundtrip time (RTT) and latency between a data center and a core point. Thus, for example, a core point may be pinged periodically (e.g., every 30 seconds) and the associated latency and packet loss data collected. Using such data, an average latency is calculated, preferably using an exponentially time-weighted average of all previous measurements and the new measurement. A similar function is used to calculate average packet loss. Using the results, a score is generated for each path between one of the data centers and the core point. The score may be generated by modifying an average latency, e.g., with a given penalty factor, that weights the average latency in a unique way to provide a download prediction. Whichever data center has the best score (representing the best-performing network connectivity for that time slice) is then associated with the core point. A full network map is created by generalizing a core point/data center data set to an
IP block/data center data set. This "unification" fills in and reduces the size of the network map and enables traffic redirection to be carried out for new local name servers.
The generated network map is then used to effect traffic redirection and load balancing. In particular, when a user's local name server makes a request for the content provider's web site (located within a set of mirrored data centers), the method preferably uses the network map to return to the local name server a list of web server IP addresses at the optimal data center. If ping data is not available for the user's local name server (of it the IP block has not been extended through unification), BGP or geo-routing can be used to make a default routing decision. Content provider-specified load balancing preferences may also be readily enforced across the data centers and/or within a particular data center. The foregoing has outlined some of the more pertinent objects and features of the present invention. These objects should be construed to be merely illustrative of some of the more prominent features and applications of the invention.
BRIEF DESCRIPTION OF THE DRAWINGS
For a more complete understanding of the present invention and the advantages thereof, reference should be made to the following Detailed Description taken in connection with the accompanying drawings, in which:
Figure 1 is an illustration of a mirrored Web site that is managed by a global traffic manager according to the present invention;
Figure 2 is a high level illustration of the components of the GTM service;
Figure 3 is a simplified illustration of a core point discovery process of the invention;
Figure 4 is a simplified illustration of how an end user request is processed by the global traffic redirection system of the present invention for a mirrored web site that has been integrated into the managed service;
Figure 5 is a flowchart describing how a map is generated by the GTM system; Figure 6 is a simplified block diagram of one implementation of the global traffic management system of the invention; and
Figure 7 is a representative traceroute generated during the core point discovery process.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT By way of brief background, it is known in the art for a Web content provider to distribute or "mirror" its Web site to ensure that the site is always available and providing acceptable performance for a global customer base. Once a Web site is distributed, global traffic management (GTM) solutions typically are used to direct users to the various mirror sites. GTM solutions use a variety of methods to determine which is the "best" mirrored site in which to direct a user. Because internet conditions are constantly changing, however, the "best" site for a particular user also varies with these conditions. The present invention is a GTM solution that maximizes availability and performance of a mirrored delivery site.
In a preferred embodiment now described, the global traffic management solution is a managed service provided by a service provider, such as a content delivery network
(CDN) service provider (CDNSP). As is well-known, a CDN is a network of geographically distributed content delivery nodes that are arranged for efficient delivery of digital content (e.g., Web content, streaming media and applications) on behalf of third party content providers. A request from a requesting end user for given content is directed to a "best" replica, where "best" usually means that the item is served to the client quickly compared to the time it would take to fetch it from the content provider origin server. Typically, a CDN is implemented as a combination of a content delivery infrastructure, a request-routing mechanism, and a distribution infrastructure. The content delivery infrastructure usually comprises a set of "surrogate" origin servers that are located at strategic locations (e.g., Internet Points of Presence, network access points, and the like) for delivering copies of content to requesting end users. The request-routing mechanism allocates servers in the content delivery infrastructure to requesting clients in a way that, for web content delivery, minimizes a given client's response time and, for streaming media delivery, provides for the highest quality. The distribution infrastructure consists of on-demand or push-based mechanisms that move content from the origin server to the surrogates. An effective CDN serves frequently-accessed content from a surrogate that is optimal for a given requesting client. In a typical CDN, a single service provider operates the request-routers, the surrogates, and the content distributors. In addition, that service provider establishes business relationships with content publishers and acts on behalf of their origin server sites to provide a distributed delivery system. A well-known commercial CDN that provides web content and media streaming is provided by Akamai Technologies, hie. of Cambridge, Massachusetts.
Thus, in one embodiment, the present invention implements a managed service for global load balancing of a content provider's mirrored Web sites. Figure 1 illustrates the basic implementation environment. In this example, the global traffic management service 100 provides global traffic management for a content provider running a pair of mirror Web sites 102 and 104 (identified by the same domain, e.g., www.akamai.com). The GTM service 100 provides improved responsiveness for end users 106 and 108 accessing the Web site by directing them to the best performing mirrored site. Figure 2 illustrates the high level technical architecture of the GTM service which, as noted above, is implemented by a CDNSP or other entity (the "managed service provider") as a managed service on behalf of content providers running mirrored Web sites. Of course, one of ordinary skill will appreciate that the inventive functionality may also be implemented in whole or in part as a product-based solution. For illustrative purposes only, and with reference to Figure 2, a preferred GTM service 200 comprises a number of components: a set of network agents 202, a set of web server agents 204, a set of map generation servers 206, and a set of name servers 208. Each such component typically is a server, such as a Pentium-based box running the Linux operating system and having application software for carrying out the functions described below, or one or more processes executing on such a machine. As will be described, data is collected by the network agents and the web server agents and delivered to the map generation servers. The map generation servers analyze the data, and at least one map server produces a map that assigns name server IP address/blocks to regions. At least one map is then uploaded to the name servers. When an end user requests access to a mirrored site domain being managed by the service, one of the name servers hands back an IP delegation that represents a "best" data center to which the user should connect.
The particular placement of the components as illustrated in the drawing is representative, and there is no requirement that any particular entity own or control a particular machine. In this embodiment, a content provider has network agents located in or near their network segment within each respective data center that hosts the mirrored Web site. Thus, for example, a pair of network agents 202a and 202b are dedicated to the content provider in data center 203a, and a pair of network agents 202c and 202d are dedicated to the content provider in data center 203b, although this is not required. These network agents preferably share the same network connection as the content provider's web servers. Thus, e.g., network agents 202a and 202b in data center 203a share network connections with the content provider's web servers 207a-c. Where the managed service provider operates a CDN, the set of network agents may be deployed in data centers in which the CDN is deployed. Of course, multiple content providers may host their web sites at a given data center and share network agents. Thus, a given network agent may collect data once for a first content provider at a given location and then share the data across all other content providers co-located in the same data center. A data center typically is located at a telecommunications facility (e.g., Exodus, Frontier Global, UUUNet, and the like) that leases space and sells connectivity to the Internet. A network agent has two (2) primary functions: running "core point" discovery
(CPD) to determine a set of "core" points, and monitoring network performance to each core point. As will be seen, the inventive system continuously pre-computes optimal maps, preferably for the user base of the entire Internet. It is able to do this effectively because the system reduces the scale of the problem by aggregating parts of the Internet and representing them with "core" points. A core point typically is representative of a set of local name servers (or other hosts) that, from the perspective of a given network location (e.g., a data center), share the point. Typically, a core point is a router on the Internet, although this is not a requirement. The information collected from the core point discovery process is fed to the map generation servers on a relatively frequent basis, e.g., one every thirty (30) seconds, to identify down routes, congestion, route changes, and other network traffic conditions that may impair or effect connectivity to a data center at which a particular mirrored site is hosted.
According to a preferred embodiment of the invention, a core (or "common") point is discovered as follows. An incremental trace route is executed from each of the set of mirrored data centers to a local name server that may be used by client to resolve a request for a replica stored at the data centers. An intersection of the trace routes at a common routing point is then identified. Thus, for example, the common routing point may be the first common point for the trace routes when viewed from the perspective of the data centers (or the last common point for the trace routes when viewed from the perspective of the local name server). The common routing point is then identified as the core point for the local name server. A core point is identified for other local name servers (or other hosts) in the same manner. Thus, a given set of mirrored data centers may have associated therewith a set of core points that are then useful in estimating the relative connectivity to the set of data centers, as is described below.
Figure 3 is a simplified diagram of the core point discovery process, in accordance with one embodiment of the invention. For purposes of example only, in Figure 3, the data center 300 corresponds to a data center located on the West Coast and the data center 302 corresponds to a data center located on the East Coast. Data center locations, of course, are merely representative. Each data center can host a mirror site for a given content provider. According to the invention, a core point 305 is discovered as follows. An incremental trace route is executed from each of a set of mirrored data centers 300, 302 to local name servers 304, 306, 308 that may be used by a client machine 310. For example, in Figure 3, the network agent (not shown) has executed a first set of traceroutes, between the data center 300 and the local name servers 304, 306 and 308, and a second set of traceroutes between the data center 302 and the local name servers 304, 306 and 308. The network path between the respective data center and the local name server(s) contain router hops, as is well known. To locate a core point, the network agent identifies a location at or close to the intersection of the trace routes at a common routing point, which is shown in Figure 3 as a core point 305. For example, the common routing point may be the first common point for the trace routes when viewed from the perspective of the data centers 300 and 302 (or the last common point for the traceroutes when viewed from the perspective of the local name server 304). The common routing point is then identified as the core point 305 for the local name server. Figure 7 illustrates a representative core point discovery process trace.
For example, if two or more different paths are traced and the same route (or routes) appears on at least a portion of all of the paths, the common routing point can lie somewhere along that common portion of the route. As noted above, generally the core point is the first common point for the trace routes when viewed from the perspective of the data centers, which is the same as the last common point for the trace routes when viewed from the perspective of the local name server.
The core point 305 need not be situated at the "exact" intersection of the trace routes. It can, for example, be located near or substantially near the intersection. It can also be located adjacent to the intersection, or it can be located at any nearby point such that measurements made to the point are representative of the measurements made at the intersection.
The network agent identifies other core points for other local name servers (or other hosts) in the same manner. Thus, a given set of mirrored data centers may have associated therewith a set having one or more core points that are then useful in estimating the relative connectivity to the set of data centers, as is described below. If network paths on the Internet are changing frequently, a network agent preferably runs core point discovery with some frequency.
As noted above, a network agent also performs the function of periodically checking the core points assigned to one or more local name servers that already have been mapped. This process is now described.
Network agents preferably make measurements to core points using Internet Control Messaging Protocol (ICMP) (or so-called "ping" packets) to evaluate such information as round trip times (RTTs), packet loss, and number of router hops. Thus, using the example in Figure 3, a given network agent periodically "pings" a core point (e.g., every 30 seconds) and collects the associated latency and packet loss. Using such data, the network agent calculates an average latency. In one embodiment, the network agent calculates average latency using an exponentially time-weighted average of all previous measurements and the new measurement. The network agent uses a similar function to calculate average packet loss. This calculation is described in more detail below. Using the results, the network agent generates a "score" for each path between one of the data centers and the core point. The score is generated, for example, by modifying an average latency with a given penalty factor that weights the average latency in a unique way to provide a download prediction. Whichever data center has the best score (representing the best-performing network connectivity for that time slice) is then associated with the core point.
Referring back to Figure 2, the web server agents 204 do test downloads to either all the web server IP addresses or to the local load balancing devices to test for availability or "aliveness" of the mirrored sites (i.e., per data center mirror or web server). Typically, a web server agent tests an object, e.g., a twenty (20) byte file available on the web server via an HTTP GET request, and check for errors and download times. In a representative embodiment, the measurements are taken periodically, e.g., every ten (10) seconds, although preferably a customer can change the timeout. An IP address is declared "dead" if more than a given percentage of the web server agents are unable to download the test object within the timeout threshold. This allows customers to set a threshold on response times so that the 'system can direct traffic away from data centers where performance suffers. The web server gents are preferably dispersed in co-location facilities, which are dispersed geographically and on a network basis. Moreover, one skilled in the art will recognize that the described functions of the web server agent could be performed by another component, such as the network agent, the map generation server, or some other server. Moreover, neither the web server agent nor its functions (such as testing the aliveness of a data center) are necessary for certain embodiments of the invention. The map generation servers 206 receive data from the network agents and the web server agents and use this data to generate maps, which describe the mirrored site that is optimal for each IP address block. In a preferred embodiment, a map is achieved by evaluating web server agent data, a time-weighted average of latency and packet loss, and BGP and geo information. Preferably, there are two (2) map generation server processes for each customer, and maps are generated periodically, e.g., every 3-5 minutes. Although not a limitation, preferably the map generation servers associate IP blocks with Internet "regions" such that a given map associates an IP block with a region number. Another data file is then used to associate region number to physical IP address, hi a representative embodiment, maps (which associate IP block to region #) are generated every few minutes and then uploaded to the name servers.
The name servers 208 hand out to the requesting end user the IP address(es) of the optimal data center. Typically, the name server response have a time to live (TTL) of about five (5) minutes, although this value may be customer-configurable. In a representative embodiment, the name servers are the same name servers used by the CDNSP to facilitate routing of end user requests to CDN content servers.
Figure 4 illustrates how a customer web site is integrated into the traffic redirection system of the present invention. In a representative embodiment, it is assumed that the customer has a distributed web site of at least two (2) or more mirrored sites. The inventive system load balances multiple subdomains/properties provided they are in the same data centers. Integration simply requires that the customer set its authoritative name server 400 to return a CNAME to the GTM name servers 408, which, thereafter, are used to resolve DNS queries to the mirrored customer site. Recursion is also disabled at the customer's authoritative name server. In operation, an end user 402 makes a request to the mirrored site using a conventional web browser or the like. The end user's local name server 404 issues a request to the authoritative name server 400 (or to a root server if needed, which returns data identifying the authoritative name server). The authoritative name server then returns the name of a name server 408 in the managed service.
The local name server then queries the name server 408 for an IP address, h response, the name server 408 responds with a set containing one or more IP addresses that are "optimal" for that given local name server and, thus, for the requesting end user. As described above, the optimal set of IP addresses is generated based on network maps created by testing the performance of representative core points on the network. The local name server selects an IP address from the "optimal" IP address list and returns this IP address to the requesting end user client browser. The browser then connects to that IP address to retrieve the desired content, e.g., the home page of the requested site.
Figure 5 is a high level flowchart illustrating how data is processed in order to create a map. Periodically (e.g., every thirty (30) seconds), the network agents ping each core point from each data center. This is step 500. At each network agent, a time-weighted average of latency, and a time-weighted average of loss, is computed. This is step 502. As will be described, the weights decay exponentially in time with a time constant that is configurable. At step 504, the data is further processed to produce a score for each data center per core point. At step 506, each core point is then associated with the name servers for which the core point was a proxy. At step 508, a map generation process goes through all of the data and decides a set of candidate data centers for each name server. At this time, any data centers that the web server agents determine are not "alive" are discarded. At step 510, the map generation process extends its decisions with respect to name servers to decisions with respect to IP block. A unifying algorithm is used to provide this functionality. This algorithm operates generally as follows. If all name servers in a BGP- geo block have agreeing ping decisions, then the decision of what data center is "optimal" is applied to the whole block. Conversely, if there is a disagreement, the block is broken up into the largest possible sub-blocks so that, in each sub-block, all the name servers agree. For any block that has no name servers, the BGP-geo candidates may be used. Referring now back to Figure 5, at step 512, the map is produced with the candidate for each block. If there are multiple candidates, the assignments are made to get as close to the load balancing targets are possible. The load balancing targets are defined, usually by the content provider, and these targets may be percentages (adding up to 100%) that breakdown the desired traffic amount by data center. This completes the map generation process.
As described above, step 502 involves generating a time-weighted average of latency and a time-weighted average of loss. More generally, this aspect of the invention provides a systematic methodology for predicting actual download times for various flow control protocols, e.g., TCP. As is known, TCP is the most commonly used flow control protocol on the Internet today. Other protocols are built on top of UDP. Neither TCP nor UDP packets can be used to monitor the state of routes on the Internet, however. According to the present invention, ICMP packets are injected into the network (e.g., by the network agents), at preferred points in time, and then routed to a suitably chosen intermediate core point. The system then looks at the behavior of the Internet induced by the ICMP probes by computing latency and packet loss. Latency is a measure of the round trip time (RTT) between the server and the core point. From maintaining a time-series of loss and latency, the system is able to predict effectively the amount of time it would take a client (that uses a name server associated with the core point) to initiate and complete a download from the server. The quality of this prediction is important for effective mapping because when a client makes a web request and there are multiple web servers from which to potentially server, it is important to be able to predict correctly which web server has the best connectivity. This is a difficult problem in general because the Internet is highly bursty and exhibits highly variable traffic conditions.
The following example illustrates how the time-weighted averages are computed in accordance with one embodiment of the invention. Assume for purposes of example only that a content provider (Figure 3) has mirror sites located at two data centers 300 (West Coast) and 302 (East Coast). The network agent "pings" the core point 305 from each data center. The network agent stores the latency and the packet loss for each measurement made. It should be understood that latency and loss parameters are merely representative of the types of signal transmission parameters that the network agent can track. Other parameters that could be measured include any parameter helpful in determining the speed, quality and/or efficiency of a network path, such as parameters indicative of outages on paths, loss in signal strength, error-control data, route changes, and the like.
For each "ping" to/from each data center to the core point, the respective network agent logs the data. Table 1 illustrates an example of the type of data that the network agent gathers over the course of measurements made every 30 seconds between the data centers and the core point. Table 1 is a table of latency measurements (data is in seconds (s)) and shows the current measurement (t=0) followed by measurements made previously.
Table 1
Figure imgf000016_0001
As Table 1 shows, based on latency, in this example the East Coast data center appears to have a smaller average latency to the core point than the West Coast data center. A time-weighted average of latency, and a time- weighted average of loss, is then computed. The weights decay exponentially in time with a time constant that is configurable (e.g., a time constant of 300 seconds). For a sequence of measurements made ( xt), where tt is the time of the ith measurement and Xf is the value measured (e.g., x,- can be the latency measurement lati or the loss measurement lossi), the time-weighted average of latency is computed as:
AverageLatency = latx e~'' ic
1=0
Assuming that the time constant C = 300 seconds, and using the data of Table 1, the average latency time series is computed as:
Using the data, the average latency for the data center 300 is computed as:
AverageLatency = ^T (31.88) i=0
To compute the exponentially time- weighted average, the network agent sums each weighted latency measurement (e.g., 31.88) and divides this sum by the sum of the weight factors (i.e., e"30 300 + e"60300 . . . etc.). Thus, the exponentially time weighed average latency for the data center 300 is computed as:
Exponentially time-weighted average = 31.88/4.0894
Exponentially time-weighted average = 7.795
As these computations show, the exponentially time-weighted average is 7.79, which differs from the computed average of 7.78. Although this difference does not appear significant in this example, it can be more significant for measurements averaged out over long periods of time, because more recent measurements will be given more weight than older measurements. The network agent determines dynamically whether core points that were once considered optimal are still so, whether core points that had been performing well (for a given time period) are now degraded, and the like. The exponentially time- weighted averaging helps also to smooth out aberrations over time in measured data and helps to indicate trends.
Using the same information, the time-weighted average latency for the East Coast data center 302 are computed in a similar manner. In addition, although not illustrated here, the network agent computes a time- weighted average of loss in the same way.
As described above, time-weighted averages are then processed to produce a score for each data center per core point. A preferred scoring function is as follows:
Score function = average latency + {[max (100, average latency)]*(penalty factor)}, where the score is a value in milliseconds. Each of the values has a millisecond unit, except for the penalty factor, which is unit-less. The value "100" is a floor or base-level value, which represents the usual round trip time required for a packet to travel between the East Coast and the West Coast. The floor is variable. The term "max" refers to selecting either the value "100" or the average latency, whichever is greater. That value is then multiplied by a penalty factor, with the result then being added to the average latency to generate the score. The penalty factor preferably is a function of the time-weighted average loss. Thus, in one illustrative embodiment, the penalty factor is some multiple of the time-weighted average loss. The multiplier may be varied, e.g., as a function of percentage of loss, with the penalty factor contribution being higher for greater packet loss.
According to the invention, it has been found that a scoring function such as described above that is based on time-weighted average latency weighted by a time- weighted average loss penalty factor affords a good approximation or "proxy" of the download time for an average size (e.g., 10Kbyte) file from the data center to an average end user. Of course, the file download time would be expected to vary as the file size is varied, but it has been found that the scoring function described above still tends to capture which data center of the mirrored set provides better performance. In other words, the absolute value of any given score is not as important as the data center-specific (e.g., East Coast vs. West Coast) values. When the scores are provided to the map generation process, the network agent associates the core point with the local name server(s) for which the core point serves as a "proxy."
The following describes a specific implementation of the global traffic redirection system as a managed service offering on behalf of content providers running mirrored web sites. Figure 6 illustrates the overall system architecture 600. As noted above, these processes typically run across multiple servers in the system. There are three logical grouping of these processes. First, the PingServer 602, PingProcessor 604, and TestPingServer 606 are running on the network agents located in the content provider's data centers. Second, the MapMaker 608, MapTester 610, and DBPusher 612 are running on another set of servers. However, these may also be run on the network agent machines if there is a shortage of servers in the network in which the global traffic management system operates. Another set of processes, called MapNote Web 614 and MapNoteDNS 616, run together on a relatively static set of machines for all customers of the system. Processes 602, 604, 608, 610, 612, 614 and 616 typically run continuously. An alert processor (not shown) detects if one or more machines on the network are non-functional and sends one or more corresponding alerts. An archive process (not shown) is used to automatically log files and other system files periodically. A file moving process (not shown) is used move data files. Each server may also run a generic process monitor (not shown), which reports data to a service provider query system.
As has been described, the global traffic management system 600 collects several pieces of data that results in a map being uploaded to the GTM name servers 615. At the beginning, Core Point Discovery (CPD) produces a list of IP addresses in a file (signifying the core points). This file is sent to each PingServer 602. Preferably, there is a PingServer process 602 running on each of the network agents that are deployed in a content provider's data center (not shown). In this embodiment, there is a pair of machines in each data center, only one PingServer process is primary. The other one is running but only takes over if the primary goes down. Each PingServer process 602 pings each of the core points approximately every 30 seconds.
Next, the ping results are sent to the PingProcessors 604. PingProcessors 604 preferably run on the same machines as the MapMakers 608, although this is not a requirement. The PingProcessors 604 process the ping results and drop the data files off for the MapMaker 608. MapMakers 608 also require data from the MapNote Web agents 614. The MapNoteWeb agents 614 are the web server agents that do test downloads from the content provider's web servers. These tests are used to determine aliveness of the webservers in the data centers as has been described.
The MapMaker 608 looks at the ping data as well as the MapNote Web data and creates a top-level map for the top-level name servers. The map is then sent to the MapTester 610 (which is usually running on the same machine). The MapTester 610 uses test ping data from the TestPingServer 606 to check a given number of (e.g., a few hundred) IP addresses in the map. This is done to make sure the map is correct, however, this processing is optional Finally, if the map passes the test, it is queued for uploading to the name servers 615.
DBPusher 612 is one other process that preferably runs on the same machines as the MapMaker process 608. This process is solely responsible for pushing out a DB file to the top-level name servers 615. This DB file completes the lookup data for the top-level name server 615. That is, the map from the MapMaker 608 contains a mapping of IP block to a virtual region number. The DB file is the data file that actually has a mapping of the region number to physical IP addresses. DBPusher 612 monitors the MapNote Web data and, in case of a failure, pushes an updated DB file to the name servers. PingServer 602 is responsible for measuring RTT and packet loss to the list of core points. The list of core points determined as follows. Preferably, there is a PingServer process running for each content provider at each data center in which a content provider is co-located. Thus, in one embodiment, the service provider deploys servers in all of a content provider's data centers. In another embodiment (not shown), ping data is shared for all customers who co-locate at a particular data center, and the GTM service provider may simply pre-deploy servers at "popular" hosting facilities to save time in integrating new customers to use the system.
The PingServer process preferably is run on each of the network agents in a data center. A leader election process (not shown) may be used to allow for the non-leader to take over if the primary fails within that same data center. PingServer includes a process that is used to ping a list of IP addresses, which the PingServer receives from a system source. Also, before the list is pinged, any IP addresses that are on a restricted list are filtered out. In particular, the primary inputs to the PingServer process are as follows:
• Restricted tree - a list of IP addresses that are not pinged.
• Routers file - the list of IP addresses that were discovered using Core Point Discovery.
The outputs of PingServer are as follows:
• Ping results - raw results of pinging IP addresses.
• Routers file - list of all IP addresses that PingServer used PingProcessor is responsible for taking the raw measurement data from PingServer and computing the time-weighted averages. The time-weighted average is computed both for RTT and packet loss measurements for the core points. The time-weighted average is computed as described above. The primary inputs to the PingProcessor process are as follows: • Ping results from PingServer
• Routers file from PingServer The outputs of PingProcessor are as follows:
• Nameserver list
• Processed ping data The MapMaker creates the map for the top-level name servers. MapMaker takes the processed ping data from PingProcessor and the aliveness data from MapNoteWeb and constructs a map. This map contains a relationship between certain IP blocks and a region number. The inputs to MapMaker are:
• Nameserver list from PingProcessor • Ping scores from PingProcessor
• BGP-Geo tree information
The outputs of MapMaker may include, for example:
• Debug map
• Map states file • Map
• Ping data MapTester is the last process before a map is uploaded to the top-level name servers. MapTester receives a candidate map from MapMaker. It then looks-up the mapping of a test IP addresses (that are pinged using TestPingServer, which is discussed more fully below). If the number of differences is below some threshold, then the map is deemed acceptable.
The map is then uploaded to the top-level name servers. The inputs to the MapTester process are:
• Debug map
• Test ping data • Stats file
• Map
• Ping data
The output of the MapTester process is:
• Map TestPingServer collects RTT and packet loss information for a small subset of IP addresses. This data is collected to ensure that the map produced by MapMaker is valid. MapTester, assuming the map is good, will then load the maps to the top-level name servers. The inputs to the TestPingServer process are:
• Restricted tree • List of IP addresses to test
The output of the TestPingServer process is:
• Test ping data
As noted above, because the MapMaker map only provides a mapping between IP block and a region number, a separate process preferably is used to provide the mapping between region number and the actual IP addresses of the webserver(s). DBPusher is responsible for processing the MapNoteWeb data and creating a DB file that is uploaded to the top-level name servers. Then, the top level name server will, after it has determined the region number for a give IP in the map, look in the corresponding DB file to find the right set of IP addresses to hand out. The input to DBPusher is: • MapNote Web data
The output to DBPusher is
• DB file for name servers - this file is pushed to the name server directly by DBPusher
MapNote Web is run on a select number of servers for all customers using the traffic management system. MapNoteWeb uses a list of target URLs (which, for example, could be stored in its configuration files) on which it performs test downloads. Preferably, these tests are simple HTTP GET requests and only time and errors are important. This data can be interpreted to determine whether or not a data center or web server is alive on dead. The download time is stored historically using a time-weighted average. Both the MapMaker and DBPusher use this data. The input to MapNoteWeb is: • Targets to measure against (stored in configuration file)
The output to MapNoteWeb is:
• Download test results
MapNoteDNS is related to MapNoteWeb, except that instead of monitoring web servers it monitors name servers. Its basic function is to do a measurement at a name server for certain names and report the time. If no answer comes back, the process will attempt to ping to determine whether it is the name server or the network that caused the failure. The inputs to MapNoteDNS are:
. • Name servers to test
• What domains to test for The output of MapNoteDNS is:
• DNS query results
Although not described in detail, various tests (that are not relevant to the present invention) may be executed to determine whether or not each of the above-described processes is running correctly. The intelligent traffic redirection system of the present invention has numerous advantages. The system continuously pre-computes optimal maps for the user-based of the entire Internet (or, if desired, a given sub-portion thereof). It is able to do this effectively because the system reduces the scale of the problem by aggregating parts of the Internet and representing them with core points. The system is also able to make different kinds of measurements depending upon the service being replicated. It combines these measurements for the core points into decisions which it then extends to the entire Internet using unification over a fallback partition of the IP address space using, e.g., BGP and geo information. The system also is unique in its ability to balance load for cost minimization.
The system is able to pre-compute an optimal mapping for the entire Internet at all points in time. In addition to being extremely fast in its ability to react to bad network conditions, it is also extremely fine-grained in its response. The system is able to detect bad server conditions quickly and is capable of interfacing with a multitude of local load balancers. By computing core points, the system makes superior measurements that mitigate the problem of intruding on firewalls and other detection mechanisms. Moreover, unlike the prior art, it can load balance load so as to minimize bandwidth costs.
The unification algorithm is advantageous in that it uses high-quality reliable information for a subspace and extends it to the entire space rather than falling back onto poorer quality information. This is achieved by utilizing the natural tree-like hierarchy of CIDR-based IP addressing in conjunction with the fact that Internet routers utilize the CIDR scheme for aggregating IP addresses to permit fast lookups. The technique enables the redirection system to extend the benefits of high quality information from a subset of the entire space of IP addresses. This is of great importance because the Internet is experiencing exponential growth. The unification algorithm affords the service provider with a means to deal intelligently with new hosts that are accessing the CDN for the first time. Current technologies do not possess a means of extending mapping decisions in this way. They either fall back to poor quality information or use a default technique, e.g., such as round robin, which essentially embodies no information whatsoever.
Predicting download times using ICMP probes and time-series techniques also provides numerous advantages. The technique does not have any restriction on the range of file sizes and download types, and it makes intelligent use of ICMP probes of different sizes to effectively estimate packet loss. The technique requires very little state for keeping track of the time-series and is able to quickly compute a new estimate using an exponentially time-weighted average of all previous measurements and the new measurement. Rather than attempting to probabilistically model TCP flows, the inventive technique provides a general method for extracting a good predictor of download times based on ICMP probes. In the preferred embodiment, the intelligent traffic redirection system is used to direct traffic to a mirrored Web site. Generalizing, the inventive system and managed service can be used in any situation where an end-user requires access to a replicated resource. As described above, the system directs end-users to the appropriate replica so that their route to the replica is good from a network standpoint and the replica is not overloaded. An "end user" may be generalized to any respective client system, machine, program or process. Thus, other uses of the system may include, without limitation, to direct caches to storage servers, to direct streaming servers to signal acquisition points, to direct logging processes to log archiving servers, to direct mail processes to mail servers, and the like.
Having thus described our invention, the following sets forth what we now claim.

Claims

1. A method of determining which of a set of content provider mirror sites should receive an end user's initial content request, comprising: identifying a set of proxy points, wherein each proxy point represents a given point in the Internet at which a trace originating from each of a set of mirror sites directed toward a given name server intersect; probing the proxy points to generate given data; generating a download predictor score for each mirror site based on the given data; identifying which mirror site provides a best download performance based on the download predictor score; associating a given name server IP address with the identified mirror site; and in response to an end user's initial content request to a given local name server, returning an IP address of the identified mirror site.
2. A method of optimizing a user's initial request to a content provider web site that is replicated at a set of mirror sites, comprising: responsive to an end user's local name server making a request to the content provider's web site, directing the request to a global load balancing service having a network map that estimates relative connectivity to the mirror sites from a set of proxy points; using the network map to return to the end user's local name server an IP address identifying an optimal mirror site at which the request may be serviced.
3. The method as described in Claim 2 wherein each core point represents an intersection of trace routes that originate from a mirror site to a local name server.
4. A method of routing a user's initial request to a content provider web site that is replicated at a set of mirror sites, comprising: responsive to an end user's local name server making a request to the content provider web site, directing the request to a global load balancing service having a network map that estimates relative connectivity to the mirror sites from a set of proxy points; determining whether the network map includes data associating the end user's local name server to one of the mirror sites; and if not, identifying a given mirror site to respond to the request using a default routing mechanism.
5. The method as described in Claim 4 wherein the default routing mechanism is BGP.
6. The method as described in Claim 4 wherein the default routing mechanism is geo-routing.
7. A method for managing global traffic redirection for a set of content providers operating mirrored sites, comprising: from each of a set of data centers that host mirrored sites, executing a given network test against each of a set of core points; generating a time-weighted average of a given metric based on data generated by executing the given network test; generating a score for each data center per core point; generating a set of candidate data centers for each of a set of name servers; associating a candidate data center to each of a set of IP address space blocks to generate a map; providing the map to a name server; and using the map to direct end user requests to a mirrored site to a given data center.
8. The method as described in Claim 7 wherein the given network test is a ping test.
9. The method as described in Claim 7 wherein the given metric is latency or packet loss.
10. The method as described in Claim 7 further including the step of discarding from the set of candidate data centers any data center that does not meet a given operating criteria.
11. The method as described in Claim 10 wherein the given operating criteria is evaluated using a file download test.
PCT/US2001/017176 2000-05-26 2001-05-25 Global load balancing across mirrored data centers WO2001093530A2 (en)

Priority Applications (4)

Application Number Priority Date Filing Date Title
CA2450394A CA2450394C (en) 2000-05-26 2001-05-25 Global load balancing across mirrored data centers
JP2001588195A JP4690628B2 (en) 2000-05-26 2001-05-25 How to determine which mirror site should receive end-user content requests
AU2001265051A AU2001265051A1 (en) 2000-05-26 2001-05-25 Global load balancing across mirrored data centers
EP01939545A EP1290853A2 (en) 2000-05-26 2001-05-25 Global load balancing across mirrored data centers

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US20801400P 2000-05-26 2000-05-26
US60/208,014 2000-05-26

Publications (2)

Publication Number Publication Date
WO2001093530A2 true WO2001093530A2 (en) 2001-12-06
WO2001093530A3 WO2001093530A3 (en) 2002-10-17

Family

ID=22772868

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2001/017176 WO2001093530A2 (en) 2000-05-26 2001-05-25 Global load balancing across mirrored data centers

Country Status (6)

Country Link
US (1) US7111061B2 (en)
EP (1) EP1290853A2 (en)
JP (1) JP4690628B2 (en)
AU (1) AU2001265051A1 (en)
CA (1) CA2450394C (en)
WO (1) WO2001093530A2 (en)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
EP1316901A2 (en) * 2001-11-30 2003-06-04 NTT DoCoMo, Inc. Content distribution system, description data distribution apparatus, content location management apparatus, data conversion apparatus, reception terminal apparatus, and content distribution method
EP1324546A1 (en) * 2001-12-28 2003-07-02 Motorola, Inc. Dynamic content delivery method and network
WO2004086720A1 (en) * 2003-03-26 2004-10-07 British Telecommunications Public Limited Company Client server model
EP2484064A2 (en) * 2009-10-02 2012-08-08 Limelight Networks, Inc. Real-time message queuing for a processing ring
US8504721B2 (en) 2000-09-26 2013-08-06 Brocade Communications Systems, Inc. Global server load balancing
US10193852B2 (en) 2002-08-07 2019-01-29 Avago Technologies International Sales Pte. Limited Canonical name (CNAME) handling for global server load balancing

Families Citing this family (157)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
AUPQ504100A0 (en) * 2000-01-11 2000-02-03 Notron (No. 325) Pty Limited A method for distribution of streamed data packets on a switched network utilising an intelligent distribution network
US6275470B1 (en) 1999-06-18 2001-08-14 Digital Island, Inc. On-demand overlay routing for computer-based communication networks
US6751191B1 (en) 1999-06-29 2004-06-15 Cisco Technology, Inc. Load sharing and redundancy scheme
US8543901B1 (en) 1999-11-01 2013-09-24 Level 3 Communications, Llc Verification of content stored in a network
US6754699B2 (en) 2000-07-19 2004-06-22 Speedera Networks, Inc. Content delivery and global traffic management network system
US6405252B1 (en) * 1999-11-22 2002-06-11 Speedera Networks, Inc. Integrated point of presence server network
US7523181B2 (en) * 1999-11-22 2009-04-21 Akamai Technologies, Inc. Method for determining metrics of a content delivery and global traffic management network
US7349979B1 (en) * 1999-12-02 2008-03-25 Cisco Technology, Inc. Method and apparatus for redirecting network traffic
US7058007B1 (en) 2000-01-18 2006-06-06 Cisco Technology, Inc. Method for a cable modem to rapidly switch to a backup CMTS
US6839829B1 (en) 2000-01-18 2005-01-04 Cisco Technology, Inc. Routing protocol based redundancy design for shared-access networks
US7886023B1 (en) 2000-01-21 2011-02-08 Cisco Technology, Inc. Method and apparatus for a minimalist approach to implementing server selection
US6996616B1 (en) * 2000-04-17 2006-02-07 Akamai Technologies, Inc. HTML delivery from edge-of-network servers in a content delivery network (CDN)
US7734815B2 (en) * 2006-09-18 2010-06-08 Akamai Technologies, Inc. Global load balancing across mirrored data centers
US6751646B1 (en) * 2000-06-01 2004-06-15 Sprint Communications Company L.P. Method and apparatus for implementing CORBA compliant name services incorporating load balancing features
AU2001280535A1 (en) * 2000-07-13 2002-01-30 Aprisma Management Technologies, Inc. Method and apparatus for monitoring and maintaining user-perceived quality of service in a communications network
US7653706B2 (en) 2000-07-19 2010-01-26 Akamai Technologies, Inc. Dynamic image delivery system
US7484002B2 (en) * 2000-08-18 2009-01-27 Akamai Technologies, Inc. Content delivery and global traffic management network system
US8060581B2 (en) * 2000-07-19 2011-11-15 Akamai Technologies, Inc. Dynamic image delivery system
US7346676B1 (en) * 2000-07-19 2008-03-18 Akamai Technologies, Inc. Load balancing service
US8341297B2 (en) 2000-07-19 2012-12-25 Akamai Technologies, Inc. Latencies and weightings in a domain name service (DNS) system
US7912978B2 (en) * 2000-07-19 2011-03-22 Akamai Technologies, Inc. Method for determining metrics of a content delivery and global traffic management network
US7725602B2 (en) * 2000-07-19 2010-05-25 Akamai Technologies, Inc. Domain name resolution using a distributed DNS network
US7155723B2 (en) * 2000-07-19 2006-12-26 Akamai Technologies, Inc. Load balancing service
US8527639B1 (en) * 2000-08-18 2013-09-03 Cisco Technology, Inc. Content server selection for accessing content in a content distribution network
US7752258B2 (en) * 2000-08-22 2010-07-06 Akamai Technologies, Inc. Dynamic content assembly on edge-of-network servers in a content delivery network
US9130954B2 (en) 2000-09-26 2015-09-08 Brocade Communications Systems, Inc. Distributed health check for global server load balancing
US7657629B1 (en) 2000-09-26 2010-02-02 Foundry Networks, Inc. Global server load balancing
US6813635B1 (en) * 2000-10-13 2004-11-02 Hewlett-Packard Development Company, L.P. System and method for distributing load among redundant independent stateful world wide web server sites
US6904602B1 (en) 2000-11-30 2005-06-07 Sprint Communications Company L.P. Method and apparatus for implementing persistence in trader services associated with a computer system
EP1388073B1 (en) * 2001-03-01 2018-01-10 Akamai Technologies, Inc. Optimal route selection in a content delivery network
US7149797B1 (en) * 2001-04-02 2006-12-12 Akamai Technologies, Inc. Content delivery network service provider (CDNSP)-managed content delivery network (CDN) for network service provider (NSP)
US20020143798A1 (en) * 2001-04-02 2002-10-03 Akamai Technologies, Inc. Highly available distributed storage system for internet content with storage site redirection
US7624184B1 (en) * 2001-06-06 2009-11-24 Cisco Technology, Inc. Methods and apparatus for managing access to data through a network device
US7881208B1 (en) 2001-06-18 2011-02-01 Cisco Technology, Inc. Gateway load balancing protocol
JP2003141006A (en) * 2001-07-17 2003-05-16 Canon Inc Communication system, communication device, communication method, storage medium and program
US6826601B2 (en) * 2001-09-06 2004-11-30 Bea Systems, Inc. Exactly one cache framework
US7113980B2 (en) 2001-09-06 2006-09-26 Bea Systems, Inc. Exactly once JMS communication
JP4160506B2 (en) 2001-09-28 2008-10-01 レヴェル 3 シーディーエヌ インターナショナル インコーポレーテッド. Configurable adaptive wide area traffic control and management
US7860964B2 (en) 2001-09-28 2010-12-28 Level 3 Communications, Llc Policy-based content delivery network selection
US7373644B2 (en) 2001-10-02 2008-05-13 Level 3 Communications, Llc Automated server replication
US20030079027A1 (en) 2001-10-18 2003-04-24 Michael Slocombe Content request routing and load balancing for content distribution networks
JP3857105B2 (en) * 2001-10-30 2006-12-13 富士通株式会社 Data transfer device
US9167036B2 (en) 2002-02-14 2015-10-20 Level 3 Communications, Llc Managed object replication and delivery
AU2003216332A1 (en) * 2002-02-21 2003-09-09 Bea Systems, Inc. System and method for message driven bean service migration
US20040006622A1 (en) * 2002-07-03 2004-01-08 Burkes Don L. Optimized process for balancing load for data mirroring
US7086061B1 (en) 2002-08-01 2006-08-01 Foundry Networks, Inc. Statistical tracking of global server load balancing for selecting the best network address from ordered list of network addresses based on a set of performance metrics
US7461147B1 (en) * 2002-08-26 2008-12-02 Netapp. Inc. Node selection within a network based on policy
US7136922B2 (en) 2002-10-15 2006-11-14 Akamai Technologies, Inc. Method and system for providing on-demand content delivery for an origin server
JP4098610B2 (en) * 2002-12-10 2008-06-11 株式会社日立製作所 Access relay device
US7233981B2 (en) * 2003-02-27 2007-06-19 Nortel Networks Limited System and method for multi-site load-balancing of encrypted traffic
GB0306973D0 (en) * 2003-03-26 2003-04-30 British Telecomm Transmitting video
US7660896B1 (en) 2003-04-15 2010-02-09 Akamai Technologies, Inc. Method of load balancing edge-enabled applications in a content delivery network (CDN)
US7251691B2 (en) * 2003-07-11 2007-07-31 International Business Machines Corporation Autonomic predictive load balancing of output transfers for two peer computers for data storage applications
GB0319251D0 (en) * 2003-08-15 2003-09-17 British Telecomm System and method for selecting data providers
US7715934B2 (en) 2003-09-19 2010-05-11 Macrovision Corporation Identification of input files using reference files associated with nodes of a sparse binary tree
US9584360B2 (en) 2003-09-29 2017-02-28 Foundry Networks, Llc Global server load balancing support for private VIP addresses
US20050097185A1 (en) * 2003-10-07 2005-05-05 Simon Gibson Localization link system
US20050089014A1 (en) * 2003-10-27 2005-04-28 Macrovision Corporation System and methods for communicating over the internet with geographically distributed devices of a decentralized network using transparent asymetric return paths
US7676561B1 (en) * 2003-12-02 2010-03-09 Sun Microsystems, Inc. Method and apparatus for adaptive load balancing
US7269603B1 (en) 2003-12-17 2007-09-11 Sprint Communications Company L.P. Enterprise naming service system and method
US7657622B1 (en) * 2003-12-23 2010-02-02 At&T Intellectual Property Ii, L.P. Unified web hosting and content distribution system and method for assuring predetermined performance levels
US7457868B1 (en) * 2003-12-30 2008-11-25 Emc Corporation Methods and apparatus for measuring network performance
US7877810B2 (en) * 2004-03-02 2011-01-25 Rovi Solutions Corporation System, method and client user interface for a copy protection service
GB0406901D0 (en) * 2004-03-26 2004-04-28 British Telecomm Transmitting recorded material
US7584301B1 (en) 2004-05-06 2009-09-01 Foundry Networks, Inc. Host-level policies for global server load balancing
US7987181B2 (en) * 2004-06-16 2011-07-26 Symantec Operating Corporation System and method for directing query traffic
US7725601B2 (en) * 2004-10-12 2010-05-25 International Business Machines Corporation Apparatus, system, and method for presenting a mapping between a namespace and a set of computing resources
US8346956B2 (en) 2004-10-29 2013-01-01 Akamai Technologies, Inc. Dynamic image delivery system
US8145908B1 (en) 2004-10-29 2012-03-27 Akamai Technologies, Inc. Web content defacement protection system
US7685312B1 (en) * 2005-02-10 2010-03-23 Sun Microsystems, Inc. Resource location by address space allocation
US7640339B1 (en) * 2005-02-14 2009-12-29 Sun Microsystems, Inc. Method and apparatus for monitoring a node in a distributed system
US7941556B2 (en) * 2005-02-23 2011-05-10 At&T Intellectual Property I, Lp Monitoring for replica placement and request distribution
US7809943B2 (en) * 2005-09-27 2010-10-05 Rovi Solutions Corporation Method and system for establishing trust in a peer-to-peer network
US8856278B2 (en) * 2005-11-16 2014-10-07 Netapp, Inc. Storage system for pervasive and mobile content
US8086722B2 (en) 2005-12-21 2011-12-27 Rovi Solutions Corporation Techniques for measuring peer-to-peer (P2P) networks
US8447837B2 (en) * 2005-12-30 2013-05-21 Akamai Technologies, Inc. Site acceleration with content prefetching enabled through customer-specific configurations
JP2007193436A (en) * 2006-01-17 2007-08-02 Fujitsu Ltd Log retrieval program, log management device, information processor and log retrieval method
US8254264B1 (en) * 2006-04-28 2012-08-28 Hewlett-Packard Development Company, L.P. Network latency estimation
US7685630B2 (en) * 2006-05-04 2010-03-23 Citrix Online, Llc Methods and systems for providing scalable authentication
EP2080105A4 (en) * 2006-09-06 2011-03-02 Akamai Tech Inc Hybrid content delivery network (cdn) and peer-to-peer (p2p) network
KR101382393B1 (en) * 2007-01-16 2014-04-09 삼성전자주식회사 Sever and simultaneous connection control method thereof
KR101109915B1 (en) * 2007-11-26 2012-02-24 에스케이플래닛 주식회사 Rich-media Transmission System and Control Method thereof
US20090144338A1 (en) * 2007-11-30 2009-06-04 Yahoo! Inc. Asynchronously replicated database system using dynamic mastership
US8489731B2 (en) * 2007-12-13 2013-07-16 Highwinds Holdings, Inc. Content delivery network with customized tracking of delivery data
US7962580B2 (en) * 2007-12-13 2011-06-14 Highwinds Holdings, Inc. Content delivery network
EP2240865B1 (en) * 2007-12-20 2020-11-04 Fast Health Corporation Method for redirecting a website upon the occurrence of a disaster or emergency event
US8015144B2 (en) 2008-02-26 2011-09-06 Microsoft Corporation Learning transportation modes from raw GPS data
US8972177B2 (en) 2008-02-26 2015-03-03 Microsoft Technology Licensing, Llc System for logging life experiences using geographic cues
US8458298B2 (en) * 2008-03-03 2013-06-04 Microsoft Corporation Failover in an internet location coordinate enhanced domain name system
US7991879B2 (en) * 2008-03-03 2011-08-02 Microsoft Corporation Internet location coordinate enhanced domain name system
US8966121B2 (en) 2008-03-03 2015-02-24 Microsoft Corporation Client-side management of domain name information
US7930427B2 (en) * 2008-03-03 2011-04-19 Microsoft Corporation Client-side load balancing
EP2101503A1 (en) * 2008-03-11 2009-09-16 British Telecommunications Public Limited Company Video coding
US8171115B2 (en) * 2008-03-18 2012-05-01 Microsoft Corporation Resource equalization for inter- and intra- data center operations
US10924573B2 (en) 2008-04-04 2021-02-16 Level 3 Communications, Llc Handling long-tail content in a content delivery network (CDN)
US9762692B2 (en) 2008-04-04 2017-09-12 Level 3 Communications, Llc Handling long-tail content in a content delivery network (CDN)
EP2274684A4 (en) 2008-04-04 2012-12-05 Level 3 Communications Llc Handling long-tail content in a content delivery network (cdn)
US8160063B2 (en) * 2008-06-09 2012-04-17 Microsoft Corporation Data center interconnect and traffic engineering
US8127020B2 (en) * 2008-08-28 2012-02-28 Red Hat, Inc. HTTP standby connection
US7962613B2 (en) * 2008-11-14 2011-06-14 At&T Intellectual Property I, Lp System and method for modifying an allocation scheme based on a resource utilization
KR101137249B1 (en) 2008-11-14 2012-04-20 에스케이플래닛 주식회사 System and method for contents delivery based on multiple Content Delivery Network providers, and content server thereof
EP2200319A1 (en) 2008-12-10 2010-06-23 BRITISH TELECOMMUNICATIONS public limited company Multiplexed video streaming
US8065433B2 (en) 2009-01-09 2011-11-22 Microsoft Corporation Hybrid butterfly cube architecture for modular data centers
US9063226B2 (en) * 2009-01-14 2015-06-23 Microsoft Technology Licensing, Llc Detecting spatial outliers in a location entity dataset
EP2219342A1 (en) 2009-02-12 2010-08-18 BRITISH TELECOMMUNICATIONS public limited company Bandwidth allocation control in multiple video streaming
US20100223364A1 (en) * 2009-02-27 2010-09-02 Yottaa Inc System and method for network traffic management and load balancing
US8209415B2 (en) * 2009-02-27 2012-06-26 Yottaa Inc System and method for computer cloud management
WO2010102084A2 (en) * 2009-03-05 2010-09-10 Coach Wei System and method for performance acceleration, data protection, disaster recovery and on-demand scaling of computer applications
US7966383B2 (en) * 2009-03-27 2011-06-21 Business Objects Software Ltd. Client-server systems and methods for accessing metadata information across a network using proxies
WO2010117623A2 (en) * 2009-03-31 2010-10-14 Coach Wei System and method for access management and security protection for network accessible computer services
US20100325182A1 (en) * 2009-06-17 2010-12-23 Ganz, An Ontario Partnership Consisting Of 2121200 Ontario Inc., And 2121812 Ontario Inc. Downloadable multimedia with access codes
US8839254B2 (en) * 2009-06-26 2014-09-16 Microsoft Corporation Precomputation for data center load balancing
US8189487B1 (en) 2009-07-28 2012-05-29 Sprint Communications Company L.P. Determination of application latency in a network node
US9009177B2 (en) 2009-09-25 2015-04-14 Microsoft Corporation Recommending points of interests in a region
KR101072983B1 (en) * 2010-01-05 2011-10-12 주식회사 유섹 Abnormal access prevention system using super grid
US20110202682A1 (en) * 2010-02-12 2011-08-18 Microsoft Corporation Network structure for data center unit interconnection
US8612134B2 (en) 2010-02-23 2013-12-17 Microsoft Corporation Mining correlation between locations using location history
US9261376B2 (en) 2010-02-24 2016-02-16 Microsoft Technology Licensing, Llc Route computation based on route-oriented vehicle trajectories
US10288433B2 (en) * 2010-02-25 2019-05-14 Microsoft Technology Licensing, Llc Map-matching for low-sampling-rate GPS trajectories
US8243960B2 (en) * 2010-03-04 2012-08-14 Bose Corporation Planar audio amplifier output inductor with current sense
US20110225121A1 (en) * 2010-03-11 2011-09-15 Yahoo! Inc. System for maintaining a distributed database using constraints
US20110225120A1 (en) * 2010-03-11 2011-09-15 Yahoo! Inc. System for maintaining a distributed database using leases
US8719198B2 (en) 2010-05-04 2014-05-06 Microsoft Corporation Collaborative location and activity recommendations
US9207993B2 (en) 2010-05-13 2015-12-08 Microsoft Technology Licensing, Llc Dynamic application placement based on cost and availability of energy in datacenters
US9183560B2 (en) 2010-05-28 2015-11-10 Daniel H. Abelow Reality alternate
US9593957B2 (en) 2010-06-04 2017-03-14 Microsoft Technology Licensing, Llc Searching similar trajectories by locations
US8549148B2 (en) 2010-10-15 2013-10-01 Brocade Communications Systems, Inc. Domain name system security extensions (DNSSEC) for global server load balancing
US8849469B2 (en) 2010-10-28 2014-09-30 Microsoft Corporation Data center system that accommodates episodic computation
US9063738B2 (en) 2010-11-22 2015-06-23 Microsoft Technology Licensing, Llc Dynamically placing computing jobs
US8825813B2 (en) 2010-12-28 2014-09-02 Microsoft Corporation Distributed network coordinate system based on network performance
US9595054B2 (en) 2011-06-27 2017-03-14 Microsoft Technology Licensing, Llc Resource management for cloud computing platforms
US9450838B2 (en) 2011-06-27 2016-09-20 Microsoft Technology Licensing, Llc Resource management for cloud computing platforms
JP5731353B2 (en) * 2011-10-21 2015-06-10 株式会社日立製作所 Cache server solving method, apparatus, and system
US9754226B2 (en) 2011-12-13 2017-09-05 Microsoft Technology Licensing, Llc Urban computing of route-oriented vehicles
US20130166188A1 (en) 2011-12-21 2013-06-27 Microsoft Corporation Determine Spatiotemporal Causal Interactions In Data
US9191163B2 (en) * 2012-03-02 2015-11-17 CMMB Vision USA Inc. Systems and methods for hybrid content delivery
US9661002B2 (en) * 2012-03-14 2017-05-23 Daniel Kaminsky Method for user authentication using DNSSEC
US10652318B2 (en) * 2012-08-13 2020-05-12 Verisign, Inc. Systems and methods for load balancing using predictive routing
US8793527B1 (en) * 2013-02-28 2014-07-29 Peaxy, Inc. Apparatus and method for handling partially inconsistent states among members of a cluster in an erratic storage network
US9998530B2 (en) * 2013-10-15 2018-06-12 Nicira, Inc. Distributed global load-balancing system for software-defined data centers
GB2525434A (en) * 2014-04-25 2015-10-28 Ibm Method and device for duplicating a data center
US10234835B2 (en) 2014-07-11 2019-03-19 Microsoft Technology Licensing, Llc Management of computing devices using modulated electricity
US9933804B2 (en) 2014-07-11 2018-04-03 Microsoft Technology Licensing, Llc Server installation as a grid condition sensor
US9729663B2 (en) 2014-09-29 2017-08-08 Limelight Networks, Inc. Dynamic/shared PMTU cache
EP3079089A1 (en) * 2015-04-08 2016-10-12 Siemens Healthcare GmbH Method for relocating medical data
US9992163B2 (en) 2015-12-14 2018-06-05 Bank Of America Corporation Multi-tiered protection platform
US9832229B2 (en) 2015-12-14 2017-11-28 Bank Of America Corporation Multi-tiered protection platform
US9832200B2 (en) 2015-12-14 2017-11-28 Bank Of America Corporation Multi-tiered protection platform
CN105704260B (en) * 2016-04-14 2019-05-21 上海牙木通讯技术有限公司 A kind of analysis method of internet traffic source whereabouts
US11240111B2 (en) 2016-06-29 2022-02-01 Nicira, Inc. Analysis of simultaneous multi-point packet capture and display of the analysis
US10270674B2 (en) 2017-05-19 2019-04-23 Akamai Technologies, Inc. Traceroutes for discovering the network path of inbound packets transmitted from a specified network node
US11005929B1 (en) 2019-01-30 2021-05-11 Cisco Technology, Inc. Dynamic data center load balancing using border gateway protocol
US11283757B2 (en) 2019-06-25 2022-03-22 Akamai Technologies, Inc. Mapping internet routing with anycast and utilizing such maps for deploying and operating anycast points of presence (PoPs)
KR102338265B1 (en) * 2019-10-07 2021-12-09 계명대학교 산학협력단 Monitoring system and method for space weather observation data in ipfs decentralized storage environment
CN111614954A (en) * 2020-05-15 2020-09-01 腾讯科技(深圳)有限公司 Index acquisition processing method and device for streaming media, computer and storage medium
CN114448990B (en) * 2021-12-23 2023-06-23 天翼云科技有限公司 Fusion CDN-based resource scheduling method, device and equipment
CN115473871B (en) * 2022-09-19 2023-08-04 广州市百果园网络科技有限公司 Domain name local resolution method and device, equipment, medium and product thereof

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1998018076A1 (en) * 1996-10-18 1998-04-30 Intervu, Inc. System and method for optimized storage and retrieval of data on a distributed computer network
EP0959601A1 (en) * 1998-05-21 1999-11-24 Sun Microsystems, Inc. A system and method for server selection for mirrored sites
WO2000014633A1 (en) * 1998-09-03 2000-03-16 Sun Microsystems, Inc. Load balancing in a network environment

Family Cites Families (26)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6003030A (en) * 1995-06-07 1999-12-14 Intervu, Inc. System and method for optimized storage and retrieval of data on a distributed computer network
CA2267953C (en) * 1996-07-25 2011-10-18 Mark Kriegsman Web serving system with primary and secondary servers
US6052718A (en) * 1997-01-07 2000-04-18 Sightpath, Inc Replica routing
US6119143A (en) * 1997-05-22 2000-09-12 International Business Machines Corporation Computer system and method for load balancing with selective control
US6578077B1 (en) * 1997-05-27 2003-06-10 Novell, Inc. Traffic monitoring tool for bandwidth management
US6263368B1 (en) * 1997-06-19 2001-07-17 Sun Microsystems, Inc. Network load balancing for multi-computer server by counting message packets to/from multi-computer server
US6112257A (en) * 1997-09-24 2000-08-29 Emc Corporation Dynamic adjustment of mirror service policy for logical volumes in a disk drive system based on collected statistics
US6360262B1 (en) * 1997-11-24 2002-03-19 International Business Machines Corporation Mapping web server objects to TCP/IP ports
JP3665460B2 (en) * 1997-12-05 2005-06-29 富士通株式会社 Route selection system, method, and recording medium by response time tuning of distributed autonomous cooperation type
US6185598B1 (en) * 1998-02-10 2001-02-06 Digital Island, Inc. Optimized network resource location
US6122666A (en) * 1998-02-23 2000-09-19 International Business Machines Corporation Method for collaborative transformation and caching of web objects in a proxy network
JPH11261580A (en) * 1998-03-12 1999-09-24 Toshiba Corp Device and method for adjusting information retrieval route on network
JPH11331256A (en) * 1998-05-20 1999-11-30 Mitsubishi Electric Corp Network connection destination changeover system
US6108703A (en) * 1998-07-14 2000-08-22 Massachusetts Institute Of Technology Global hosting system
US6295275B1 (en) 1998-08-19 2001-09-25 Mci Communications Corporation Dynamic route generation for real-time network restoration using pre-plan route generation methodology
US6546423B1 (en) * 1998-10-22 2003-04-08 At&T Corp. System and method for network load balancing
US6304913B1 (en) * 1998-11-09 2001-10-16 Telefonaktiebolaget L M Ericsson (Publ) Internet system and method for selecting a closest server from a plurality of alternative servers
US6442140B1 (en) * 1999-01-04 2002-08-27 3Com Corporation Method for automatic setup of missing RM cell count parameter CRM in an ATM traffic management descriptor
US6587878B1 (en) * 1999-05-12 2003-07-01 International Business Machines Corporation System, method, and program for measuring performance in a network system
US6295015B1 (en) * 1999-09-01 2001-09-25 Zenith Electronics Corporation Reference generator for demodulator calibration
US6484143B1 (en) * 1999-11-22 2002-11-19 Speedera Networks, Inc. User device and system for traffic management and content distribution over a world wide area network
US6405252B1 (en) * 1999-11-22 2002-06-11 Speedera Networks, Inc. Integrated point of presence server network
US6769030B1 (en) 2000-02-07 2004-07-27 International Business Machines Corporation Method and apparatus to evaluate and measure the optimal network packet size for file transfer in high-speed networks
US6820133B1 (en) * 2000-02-07 2004-11-16 Netli, Inc. System and method for high-performance delivery of web content using high-performance communications protocol between the first and second specialized intermediate nodes to optimize a measure of communications performance between the source and the destination
US6684250B2 (en) * 2000-04-03 2004-01-27 Quova, Inc. Method and apparatus for estimating a geographic location of a networked entity
US6764030B2 (en) * 2001-01-12 2004-07-20 Diamond Power International, Inc. Sootblower nozzle assembly with an improved downstream nozzle

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO1998018076A1 (en) * 1996-10-18 1998-04-30 Intervu, Inc. System and method for optimized storage and retrieval of data on a distributed computer network
EP0959601A1 (en) * 1998-05-21 1999-11-24 Sun Microsystems, Inc. A system and method for server selection for mirrored sites
WO2000014633A1 (en) * 1998-09-03 2000-03-16 Sun Microsystems, Inc. Load balancing in a network environment

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
GUYTON J D ET AL: "LOCATING NEARBY COPIES OF REPLICATED INTERNET SERVERS" COMPUTER COMMUNICATIONS REVIEW, ASSOCIATION FOR COMPUTING MACHINERY. NEW YORK, US, vol. 25, no. 4, 1 October 1995 (1995-10-01), pages 288-298, XP000541664 ISSN: 0146-4833 *

Cited By (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9479574B2 (en) 2000-09-26 2016-10-25 Brocade Communications Systems, Inc. Global server load balancing
US8504721B2 (en) 2000-09-26 2013-08-06 Brocade Communications Systems, Inc. Global server load balancing
US9225775B2 (en) 2000-09-26 2015-12-29 Brocade Communications Systems, Inc. Global server load balancing
EP1316901A2 (en) * 2001-11-30 2003-06-04 NTT DoCoMo, Inc. Content distribution system, description data distribution apparatus, content location management apparatus, data conversion apparatus, reception terminal apparatus, and content distribution method
EP1316901A3 (en) * 2001-11-30 2005-09-28 NTT DoCoMo, Inc. Content distribution system, description data distribution apparatus, content location management apparatus, data conversion apparatus, reception terminal apparatus, and content distribution method
US6952712B2 (en) 2001-11-30 2005-10-04 Ntt Docomo, Inc. Method and apparatus for distributing content data over a network
EP1675027A1 (en) * 2001-11-30 2006-06-28 NTT DoCoMo, Inc. Content distribution system, description data distribution apparatus, content location management apparatus, data conversion apparatus, reception terminal apparatus, and content distribution method
EP1324546A1 (en) * 2001-12-28 2003-07-02 Motorola, Inc. Dynamic content delivery method and network
WO2003065659A1 (en) * 2001-12-28 2003-08-07 Motorola Inc Dynamic content allocation/delivery mechanism
US11095603B2 (en) 2002-08-07 2021-08-17 Avago Technologies International Sales Pte. Limited Canonical name (CNAME) handling for global server load balancing
US10193852B2 (en) 2002-08-07 2019-01-29 Avago Technologies International Sales Pte. Limited Canonical name (CNAME) handling for global server load balancing
WO2004086720A1 (en) * 2003-03-26 2004-10-07 British Telecommunications Public Limited Company Client server model
EP2484064A4 (en) * 2009-10-02 2015-04-22 Limelight Networks Inc Real-time message queuing for a processing ring
EP2484064A2 (en) * 2009-10-02 2012-08-08 Limelight Networks, Inc. Real-time message queuing for a processing ring

Also Published As

Publication number Publication date
AU2001265051A1 (en) 2001-12-11
CA2450394A1 (en) 2001-12-06
WO2001093530A3 (en) 2002-10-17
EP1290853A2 (en) 2003-03-12
JP4690628B2 (en) 2011-06-01
JP2004507128A (en) 2004-03-04
CA2450394C (en) 2011-07-19
US7111061B2 (en) 2006-09-19
US20020129134A1 (en) 2002-09-12

Similar Documents

Publication Publication Date Title
CA2450394C (en) Global load balancing across mirrored data centers
US7734815B2 (en) Global load balancing across mirrored data centers
US7096263B2 (en) Method for predicting file download time from mirrored data centers in a global computer network
US7251688B2 (en) Method for generating a network map
US7028083B2 (en) Method for extending a network map
US7725602B2 (en) Domain name resolution using a distributed DNS network
Madhyastha et al. iPlane: An information plane for distributed services
US7657629B1 (en) Global server load balancing
Colajanni et al. Analysis of task assignment policies in scalable distributed Web-server systems
US7447798B2 (en) Methods and systems for providing dynamic domain name system for inbound route control
US7441045B2 (en) Method and system for balancing load distribution on a wide area network
US7020698B2 (en) System and method for locating a closest server in response to a client domain name request
US6795858B1 (en) Method and apparatus for metric based server selection
Hohlfeld et al. Characterizing a meta-CDN
EP2385656B1 (en) Method and system for controlling data communication within a network
AU2010202981B2 (en) Global load balancing across mirrored data centers
WO2002006961A2 (en) Method for determining metrics of a content delivery
Tomic et al. Implementation and efficiency analysis of composite DNS-metric for dynamic server selection
Tomić et al. Dynamic Server Selection by Using a Client Side Composite DNS-Metric
Swain et al. A study of data source selection in distributed web server system
Bryhni et al. Load balancing techniques for scalable web servers

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A2

Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EC EE ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NO NZ PL PT RO RU SD SE SG SI SK SL TJ TM TR TT TZ UA UG UZ VN YU ZA ZW

AL Designated countries for regional patents

Kind code of ref document: A2

Designated state(s): GH GM KE LS MW MZ SD SL SZ TZ UG ZW AM AZ BY KG KZ MD RU TJ TM AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE TR BF BJ CF CG CI CM GA GN GW ML MR NE SN TD TG

121 Ep: the epo has been informed by wipo that ep was designated in this application
DFPE Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101)
AK Designated states

Kind code of ref document: A3

Designated state(s): AE AG AL AM AT AU AZ BA BB BG BR BY BZ CA CH CN CO CR CU CZ DE DK DM DZ EC EE ES FI GB GD GE GH GM HR HU ID IL IN IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MA MD MG MK MN MW MX MZ NO NZ PL PT RO RU SD SE SG SI SK SL TJ TM TR TT TZ UA UG UZ VN YU ZA ZW

AL Designated countries for regional patents

Kind code of ref document: A3

Designated state(s): GH GM KE LS MW MZ SD SL SZ TZ UG ZW AM AZ BY KG KZ MD RU TJ TM AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE TR BF BJ CF CG CI CM GA GN GW ML MR NE SN TD TG

ENP Entry into the national phase

Ref country code: JP

Ref document number: 2001 588195

Kind code of ref document: A

Format of ref document f/p: F

WWE Wipo information: entry into national phase

Ref document number: 2001265051

Country of ref document: AU

REG Reference to national code

Ref country code: DE

Ref legal event code: 8642

WWE Wipo information: entry into national phase

Ref document number: 2001939545

Country of ref document: EP

WWP Wipo information: published in national office

Ref document number: 2001939545

Country of ref document: EP

WWE Wipo information: entry into national phase

Ref document number: 2450394

Country of ref document: CA