CN100471112C - Memory-network memory-magnetic disc high speed reliable storage system and its reading/writing method - Google Patents
Memory-network memory-magnetic disc high speed reliable storage system and its reading/writing method Download PDFInfo
- Publication number
- CN100471112C CN100471112C CNB2004100034610A CN200410003461A CN100471112C CN 100471112 C CN100471112 C CN 100471112C CN B2004100034610 A CNB2004100034610 A CN B2004100034610A CN 200410003461 A CN200410003461 A CN 200410003461A CN 100471112 C CN100471112 C CN 100471112C
- Authority
- CN
- China
- Prior art keywords
- lnd
- node computer
- write
- data
- memory
- Prior art date
- Legal status (The legal status 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 status listed.)
- Expired - Fee Related
Links
Images
Abstract
Through network, the method simulates local hard disk by using memory in remote machine in order to raise data processing efficiency and reliability of system data so that a distributed memory system in high performance is constructed as a short form LND. LND includes: a client end i.e. a computer in busy node; multiple servers, which are computers at more free node to provide memory for local use; the said client end and the said computers are connected through Ethernet. LND system constructs a virtual disk from memories at free nodes, and the client end maps the virtual disk to a local directory. Thus, reading/writing the directory is converted to reading/writing memory at remote side. The invention realizes communication between nodes through group communication protocol. Synchronization reading/writing performance of the invention is better than SCSI hard disk system.
Description
Technical field
The reading/writing method of internal memory-network internal storage-disk high speed reliable memory LND of system belongs to the computer memory system technical field of structures.
Background technology
Improving the reliability of computer system application and carrying out efficient is the important topic of computer realm research, the technical problem that wherein involves concentrates on two aspects: at first be the various measures that improve application reliability, this comprises different technologies intensive at IO and compute-intensive applications; Another then is the readwrite performance problem of computer memory system, comprises the hierarchical structure design of storage inside system, the development of various memory device.But in traditional computer application environment, the raising of this two aspects performance always exists conflicting relation, and this is relevant with the deficiency of current computer system storage hierarchy to a great extent.
CPU, internal memory, disk are three main levels that constitute current computer processing and storage system, but development at all levels is unbalanced, and according to the Moore law, cpu performance and memory size just can be doubled in per 18 months.But the capital equipment as external memory---hard disc, because the substantive characteristics of mechanical movement causes the raising of performance very limited, annual raising about 7%, this contradiction is outstanding day by day, and (access time of hard disk is mainly located by data, tracking and rotational latency, reading and writing data are formed, and wherein tracking refers to magnetic head with rotation and moves to the relevant position of hard disk so that carry out data access.Because the two is mechanically actuated operation, its performance and data location differ two more than the order of magnitude with accessing operation, and this has just hindered the raising of disk access performance).
For relaxing the performance difference between storage hierarchy, the most general a solution is that a part of internal memory of this machine is divided out, as the disk buffering memory, thereby reduces the influence of the delay of disk access to systematic function.Therefore, application program to the write operation of disk in fact being converted the write operation of buffer storage and is in time write disk with these data by operating system according to system running state.Disk write operation repeatedly can be simplified to once-through operation like this, thereby reduce magnetic disc access times, improve overall performance.This write operation is commonly referred to as asynchronous write.
In case but this scheme has been brought negative effect to the reliability of system---computer breaks down, and the data of buffer storage must be lost, and this has just brought the contradiction between data reliability and the performance.Have to any sacrifice in performance and synchronous write by at a slow speed is (different with asynchronous write as database application, synchronous write is not used buffer storage, but just calculate complete operation after directly data being write the relevant position of hard disk, therefore performance is very low, normally one of percentage of asynchronous write or one thousandth) daily record data of key is write disk.
We notice that when processor speed promoted, the bandwidth of LAN and transmission rate were also in lifting at full speed.Therefore, can pass through network, utilize the internal memory of the machine of far-end to simulate this machine internal memory, improve performance.Have following some advantage in the far-end: at first, its access speed is faster than this machine hard disk.Present network transmission speed has reached 100Mbits/sec, the network of ATM155Mbits/sec on OC-3 is also very common, even the myrinet of 1000Mbits/sec or gigabit Ethernet also occur, internal storage access is very fast again, and random read-write does not have long this fatal weakness of hard disk seek time arbitrarily.As long as the size that guarantees each transmission block suitably, surpass the hard disk of this machine on fully can assurance speed.
This invention is exactly at the problem in the current computer storage architecture, proposes to utilize the inner idle node internal memory of LAN to simulate local resource, to reach the purpose that improves data-handling efficiency and system data reliability simultaneously.
Summary of the invention
The invention reside in provides a kind of reading/writing method based on internal memory-network internal storage-disk high speed reliable memory LND of system, so that make full use of the free memory resource of each computer node under the local area network (LAN) condition, to improve the disk access performance and the application reliability of the busy node of network simultaneously, solve the problem that exists in the current computer storage organization level.
The LND system makes full use of the network bandwidth of local area network (LAN) inside and the memory source on the idle node, is I/O intensive applications program---and handle service as Web Server and db transaction, a high performance distributed storage level is provided.Its main innovative point is as follows:
1. utilize internal memory on the inner idle node of local area network (LAN) as the expansion of storage resources.
2. propose to utilize group communication protocol to realize the new thought of communicating by letter between node, improved the efficient and the reliability of system effectively.
3. having adopted the virtual device notion---the LND system is made into a virtual disk equipment with the memory group of idle node, the busy node computer that uses LND like this is after arriving certain local catalogue to this virtual disk device map, the user can be automatically converted to read-write to the far-end internal memory to the read-write of this catalogue, and is fully transparent for application program.
The method of the invention is characterised in that:
It utilizes the internal memory of local area network (LAN) interior one or many idle node computers to form a virtual disk equipment, and the busy node computer in the same local area network (LAN) is this virtual disk device map certain file directory to its this locality, and make busy node computer be automatically converted to read-write to idle node calculator memory to the read-write of this catalogue, it contains successively and has the following steps:
(1), the LND_Server process of startup one or many idle node computers, distribute an internal memory in this locality according to predefined size, set up thereon and have the file system of default form, and add the affiliated network multicast address of predefined idle node computer respectively, wait for the write request that busy node computer is sent then;
(2), start the LND_Client of busy node computer, it sends register requirement according to predefined configuration to each LND_Server, obtains this or a plurality of idle node computer externally provides shared memory size information, carries out authentication;
(3), LND_Server carries out self initialization;
(4), carry out this virtual disk device initialize, form a file directory with LND_Client;
(5), then the LND_Client use is operated this virtual disk equipment with the on all four system call of generic-document system;
(6), when LND_Client receives the data write order, destination address is carried out addressing, find network multicast address with the destination address correspondence, carry out transfer of data by one or more the idle node computer under self-defining reliable multicast protocol and this network multicast address then, to write data, write this machine hard disk as local backup with the asynchronous read and write pattern simultaneously;
(7), after the affirmation information of receiving idle node computer, LND_Client returns,
Self-defining reliable multicast protocol wherein, it contains successively and has the following steps:
At busy this end of node computer:
(1), LND_Client receives after the write order, searches the sector that writes and belongs to which network multicast address, the network address of the idle node computer of receiving serial number, last time affirmation, write address information and send to this multicast address together with data content;
(2), again data content is write local hard drive and back up, and wait for idle node computer affirmation;
(3), receive the affirmation that any one idle node computer of belonging to this multicast address sends to this serial number after, LND_Client returns;
Otherwise retransmit (4);
(5), also do not receive acknowledge message, just local hard drive is write and change synchronous mode into if retransmit certain number of times;
At idle node computer terminal:
(1), behind every LND_Server process initiation, adds the multicast address of organizing under the idle node computer oneself, and begin to receive data;
(2), if the serial number that receives is to receive serial number+1 last time, directly sends and confirm this serial number to busy node computer;
(3), if the serial number that receives data is greater than receiving serial number+1 last time, the network address of just receiving the idle node computer of affirmation in the data that receive specified last time sends request, asked for from last time and received the data of serial number+1 in this scope of serial number-1 that receives data, returned affirmation to busy node computer at last.
The present invention compares with the synchronous write performance of the scsi device of using as memory, and under the JA(junction ambient) of 100M Ethernet, LND is 3-8 a times that SCSI gets write performance; Under the 10M ethernet environment, the former also be the latter 1.5-3 doubly.
Description of drawings
Fig. 1, the structure chart of system of the present invention, i.e. LND system model.
Fig. 2, the FB(flow block) of system of the present invention.
Embodiment:
The LND system is made up of a busy node computer and a plurality of idle node computer, and two class computers are interconnected by 100,000,000 or ten mbit ethernets.Application program of user operates in busy node computer, obtains the high-performance and the high availability of disk synchronous writing operation by the virtual disk equipment on it of visiting.
The virtual disk equipment of LND is to offer the user's with a virtual block device form.The advantage of block device is that interface is transparent fully, and the user can carry out subregion to it to the operation of block device and identical to the operation of common hard disk, format, and on top set up file system arbitrarily.Transparent like this interface can guarantee that application program of user does not add all advantages that any change just can obtain the LND system.
The address space of whole LND equipment is to be determined by the memory size that the configuration of system and idle node computer are contributed.When system needs higher availability, can select that two or more idle node computers are formed mirror image pattern and back up mutually, and when system needs more space and performance, then can remove mirror image pattern, raising speed and capacity.
The running environment of whole LND system is made of computer, network and LND software three.The network of being supported is the 10M/100M/1000M Ethernet; The computer structure of supporting is counted and from 2 to 128 is not waited; Simultaneously LND software is divided into two kinds of client and server ends, is called LND_Client and LND_Server.
A LND system configuration the simplest can be made of two All-China Federation of Taiwan Compatriots's net computers, and wherein an idle node computer run LND_Server provides free memory; The application that busy node computer run is had relatively high expectations to disk performance, it utilizes LND_Client to use these internal memories.The running of whole LND system is as follows:
At first start LND_Server, it distributes an internal memory according to predefined size in this locality, and sets up file system (default form is ext3) thereon, waits for the request that client is sent then.
After LND_Client starts,, send register requirement, obtain the information of sharing out on this node such as memory size, and carry out authentication to LND_Server according to predefined configuration.
LND_Server carries out the initialization of self, sets up related data structure.
LND_Client carries out the initialization of virtual disk equipment, and it is exactly to have increased a file directory In the view of this machine user, will be automatically converted to the visit of LND_Server being gone up internal memory to the visit of this catalogue.
Concrete browsing process is as follows:
Application program on the at first busy node computer is used with the on all four system call of generic-document system LND virtual disk equipment is operated.
When the LND virtual disk is received the order that the user sends, at first destination address is carried out addressing, find the network address with the idle node computer of destination address correspondence.
Secondly LND_Client is sent to corresponding idle node computer by our self-defining reliable multicast protocol with data, adopts the asynchronous write pattern to write the hard disk of this machine simultaneously.
After the affirmation information of receiving idle node computer, system call is returned.
Certainly under this easy configuration, the reliability of system can not get guaranteeing: in a single day idle node computer breaks down, and data also can be lost.The LND system provides the mode of grouping mirror image and local backup to realize the reliability of system.
Mirror image is the most simple and practical redundant fashion.Under this pattern, we are divided into the n group to all idle node computers earlier, and each group is mapped to the different address spaces of virtual disk equipment respectively.Suppose that each group has m platform machine, then this m platform machine is done mirror back-up mutually.The machine that it is pointed out that top indication is actually notion in logic, and m*n platform machine might not be arranged, and the machine that free memory is many can move a plurality of LND Server processes.Certainly, machine on the same group must be accomplished to distribute physically, could guarantee reliability like this.
Like this, LND_Client just changes into simultaneously m write operation that is in idle node computer on the same group the write operation of LND_Server, promptly carries out data backup.For the network that reduces write data transmits expense, LND has utilized network Multicast mechanism, on insecure udp protocol basis, has set up the reliable group communication protocol of oneself, finishes data and transmits.Because to send be " walk abreast " to data in Multicast, need not on the same group node is sent respectively, so reduced transmitting time.
Another advantage of mirror image is that availability is higher.After any machine in the group broke down, system still can continue operation under the prerequisite that does not reduce performance.Communication protocol has guaranteed the consistency of mirror image data in most times, and guarantees that the every blocks of data on the LND equipment is kept in the internal memory of two machines at least at any time.Such reliability is enough to satisfy the requirement of most application programs.
Table 1 has been listed the corresponding fault-tolerant ability of fault type and LND system that system may run into.First three plants the reliability that fault does not influence data, and reconstitution time is very short.For the 4th kind of situation, suppose that every machine on average collapses once weekly, every group has two machines, and so busy node computer and all idle node computers coexist in same minute together that the time interval of collapse is approximately 2,000,000 years, can think that this situation can not occur.
Table 1.LND Fault Tolerance table
| Fault type | Can work on | Performance impact | Reconstitution time | Whether loss of data |
| Idle node computer glitch of part or subnetwork fault | Energy | Do not influence | Do not have | Not |
| Whole idle node computer glitchs or overall network fault | Energy | Be reduced to the local hard drive performance | Hard disk-network-memory copying time | Not |
| Busy node computer crashes | Not | N/A | Internal memory-network-disk copy time | Not |
| All machines collapse simultaneously | Not | N/A | Hard disk-network-internal memory | Have |
| Burst | Copy time |
LND_Client, it has handled all input and output at virtual disk equipment, and these requests is sent on the LND_Server of idle node computer.Adopted our self-designed simple group communication protocol simultaneously.
In order to narrate conveniently, suppose that the idle node computer in the LND system is total to the n group, and every group computer number is m.Each group has a network multicast address.LND_Client receives after the write order, at first find out the sector that writes and belong to which multicast address, serial number seq, received the network address of the busy node computer (being defined as lastserver) of affirmation last time, write address information and data content and send to this multicast address.For each group, serial number is unique and increases progressively that it is identifying this and is organizing write operation each time.Then, with content asynchronous write local hard drive, and wait for the affirmation of any one idle node computer.After receiving the affirmation to this serial number seq that any one idle node computer sends, LND_Client can successfully return.Do not receive confirmation yet if surpass certain hour, then think packet loss and data retransmission.Number of retransmissions surpasses certain number of times and does not receive that yet any affirmation thinks that then this is organized all computers and all breaks down, and local hard drive is write change synchronous mode into, treats to return after this secondary data writes hard disk.
At idle node computer terminal, at first add the multicast address of own affiliated group after every the LND_Server process initiation, and begin to receive data.Received serial number lastSeq last time to add 1 if data are accepted the serial number seq of data, then expression does not have packet loss, directly this serial number is confirmed to beam back client.If seq〉lastSeq+1, so just ask for these unreceived data from lastSeq+1 to seq-1 to LastServer.Return affirmation at last.
The source program of LND is made up of two parts.LND_Client and LND_Server.Wherein LND_Client operates in busy node computer terminal, and LND_Server operates in idle node computer terminal.
LND_Client comprises two parts: the one, and device driver can be compiled into module and add the internal memory execution dynamically.Wherein realized the necessary lnd_request of device driver (), lnd_release, lnd_ioctl (), lnd_open (), lnd_fops (), lnd_init (), init_module (), clean_module ().It will catch user's operations such as read-write control all to virtual disk equipment, and send it to LND_Daemon, i.e. second portion.This part is the client process that operates in user's attitude, and owing to the function that can call at kernel mode is very limited, function realizes more complicated, also makes mistakes easily, so LND handle and network function associated have realized in LND_Daemon.The change that needs when this can reduce operating system transplantation to greatest extent, and avoid modification to the operating system kernel code.
We have realized group communication protocol with socket, and have realized following function:
Restore_lnd_server () function: the data in the server of collapse are recovered;
Lnd_cleaner () function: the redundant removing;
Lnd_malloc () function: storage allocation, if make a mistake then return error code;
Lnd_client_sighandler () function: catch the SIGINT signal, carry out removing work;
LND_Server is the process that operates in idle node computer terminal, is in user's attitude during operation.Wherein topmost function is exactly handle_request (), mainly handles following incident:
NRD_REQ_BLOCK_WRITE: the write request of client
NRD_REQ_BLOCK_READ: the read request of client
NRD_REQ_STATS: the internal memory operating position of obtaining server end
NRD_REQ_STOP_SERVER: stop the network internal storage service.
Three PII400, in save as 256,000,000, dispose the SCSI hard disk, built the LND system on the node computer with 100M network ethernet interconnect, three nodes all move Redhat Linux 6.2.Wherein one is moved LND_Client, in addition two idle node computers of conduct (operation LND_Server adopts the server mirroring mode).
On LND_Client, move Lmbench (C.Staelin, lmbench:Portable Tools for Performance Analysis, USENIX Annual Technical Conference.1996) test program (in order to the piece write performance of testing equipment), in the loading catalogue of the virtual disk equipment of local disk catalogue and LND, carry out the synchronous write operation of 10000 secondary data pieces respectively, the size of data block is respectively 1K, 2K, 4K, 8K, (table 2), thereby the synchronous write performance of comparison scsi device and virtual disk equipment.As can be seen from the table, under the JA(junction ambient) of 100M Ethernet, the write performance of LND is 3～8 times of SCSI; Even and if under the 10M ethernet environment, the former also is 1.5～3 times of the latter.
The synchronous write performance of table 2.LND equipment
| Equipment/file system data block size | 512Byte | 1024Byte | 2048Byte | 4096Byte |
| LND (100M) synchronous write performance | 0.7856 | 0.7929 | 0.7905 | 0.7922 |
| LND (10M) synchronous write performance | 0.3511 | 0.3545 | 0.3577 | 0.3599 |
| SCSI synchronous write performance | 0.0912 | 0.13 | 0.17 | 0.21 |
Claims (1)
1. the reading/writing method of internal memory-network internal storage-disk high speed reliable memory LND of system is characterized in that:
It utilizes the internal memory of local area network (LAN) interior one or many idle node computers to form a virtual disk equipment, and the busy node computer in the same local area network (LAN) is this virtual disk device map certain file directory to its this locality, and make busy node computer be automatically converted to read-write to idle node calculator memory to the read-write of this catalogue, it contains successively and has the following steps:
(1), the LND_Server process of startup one or many idle node computers, distribute an internal memory in this locality according to predefined size, set up thereon and have the file system of default form, and add the affiliated network multicast address of predefined idle node computer respectively, wait for the write request that busy node computer is sent then;
(2), start the LND_Client of busy node computer, it sends register requirement according to predefined configuration to each LND_Server, obtains this or a plurality of idle node computer externally provides shared memory size information at this, carries out authentication;
(3), LND_Server carries out self initialization;
(4), carry out this virtual disk device initialize, form a file directory with LND_Client;
(5), then the LND_Client use is operated this virtual disk equipment with the on all four system call of generic-document system;
(6), when LND_Client receives the data write order, destination address is carried out addressing, find network multicast address with the destination address correspondence, carry out transfer of data by one or more the idle node computer under self-defining reliable multicast protocol and this network multicast address then, to write data, write this machine hard disk as local backup with the asynchronous read and write pattern simultaneously;
(7), after the affirmation information of receiving idle node computer, LND_Client returns,
Self-defining reliable multicast protocol wherein, it contains successively and has the following steps:
At busy this end of node computer:
(1), LND_Client receives after the write order, searches the sector that writes and belongs to which network multicast address, the network address of the idle node computer of receiving serial number, last time affirmation, write address information and send to this multicast address together with data content;
(2), again data content is write local hard drive and back up, and wait for idle node computer affirmation;
(3), receive the affirmation that any one idle node computer of belonging to this multicast address sends to this serial number after, LND_Client returns;
Otherwise retransmit (4);
(5), also do not receive acknowledge message, just local hard drive is write and change synchronous mode into if retransmit certain number of times;
At idle node computer terminal:
(1), behind every LND_Server process initiation, adds the multicast address of organizing under the idle node computer oneself, and begin to receive data;
(2), be to receive serial number+1 last time if receive the serial number of data, then directly send and confirm this serial number to busy node computer;
(3), if the serial number that receives data is greater than receiving serial number+1 last time, the network address of just receiving the idle node computer of affirmation in the data that receive specified last time sends request, asked for from last time and received the data of serial number+1 in this scope of serial number-1 that receives data, returned affirmation to busy node computer at last.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2004100034610A CN100471112C (en) | 2004-03-26 | 2004-03-26 | Memory-network memory-magnetic disc high speed reliable storage system and its reading/writing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2004100034610A CN100471112C (en) | 2004-03-26 | 2004-03-26 | Memory-network memory-magnetic disc high speed reliable storage system and its reading/writing method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1564517A CN1564517A (en) | 2005-01-12 |
| CN100471112C true CN100471112C (en) | 2009-03-18 |
Family
ID=34477601
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2004100034610A Expired - Fee Related CN100471112C (en) | 2004-03-26 | 2004-03-26 | Memory-network memory-magnetic disc high speed reliable storage system and its reading/writing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN100471112C (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100403246C (en) * | 2005-06-30 | 2008-07-16 | 技嘉科技股份有限公司 | Device for simulating rigid disk and method thereof |
| CN100437522C (en) * | 2005-09-09 | 2008-11-26 | 中国科学院计算技术研究所 | Long-distance inner server and its implementing method |
| CN100458676C (en) * | 2005-09-30 | 2009-02-04 | 英业达股份有限公司 | System and method for accelerating data read and write speed |
| CN100405304C (en) * | 2006-07-27 | 2008-07-23 | 清华大学 | Method for realizing high speed solid storage device based on storage region network |
| CN101741866B (en) * | 2010-02-01 | 2012-12-12 | 浪潮(北京)电子信息产业有限公司 | On-line storage system and method |
| CN102902487B (en) * | 2011-07-25 | 2016-06-08 | 辽宁星云互联科技有限公司 | Hard disc data access system |
| CN102437925B (en) * | 2011-12-01 | 2018-03-13 | 中兴通讯股份有限公司 | Data back up method, apparatus and system in distributed system |
| CN102662610B (en) * | 2012-04-20 | 2016-05-11 | 中国科学院遥感应用研究所 | A kind of Remote Sensing Image Management System and method |
| CN105468718B (en) * | 2015-11-18 | 2020-09-08 | 腾讯科技（深圳）有限公司 | Data consistency processing method, device and system |
| CN106648464B (en) * | 2016-12-22 | 2020-01-21 | 柏域信息科技（上海）有限公司 | Multi-node mixed block cache data reading and writing method and system based on cloud storage |
| JP6689242B2 (en) * | 2017-09-19 | 2020-04-28 | 株式会社東芝 | Magnetic disk device and magnetic head control method |
-
2004
- 2004-03-26 CN CNB2004100034610A patent/CN100471112C/en not_active Expired - Fee Related
Non-Patent Citations (3)
| Title |
|---|
| . . |
| . .;Challenge and Methods of Synchronous Disk I/O. 靳超,郑纬民,汪东升,毛昀.Tsinghua Science and Technology,Vol.9 No.1. 2004 * |
| Challenge and Methods of Synchronous Disk I/O. 靳超,郑纬民,汪东升,毛昀.Tsinghua Science and Technology,Vol.9 No.1. 2004 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1564517A (en) | 2005-01-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US20220188276A1 (en) | Metadata journal in a distributed storage system | |
| US7275050B2 (en) | Storage system, a method of file data backup and method of copying of file data | |
| EP0130349B1 (en) | A method for the replacement of blocks of information and its use in a data processing system | |
| CN100405304C (en) | Method for realizing high speed solid storage device based on storage region network | |
| US6678788B1 (en) | Data type and topological data categorization and ordering for a mass storage system | |
| EP1864221B1 (en) | Network-accessible memory element, network-accessible memory module, network storage system and storage area network accelerator | |
| US6691209B1 (en) | Topological data categorization and formatting for a mass storage system | |
| US6553389B1 (en) | Resource availability determination mechanism for distributed data storage system | |
| US20090034377A1 (en) | System and method for efficient updates of sequential block storage | |
| JP2003162439A (en) | Storage system and control method therefor | |
| WO2012172553A1 (en) | Replication techniques with content addressable storage | |
| CN1723434A (en) | Apparatus and method for a scalable network attach storage system | |
| US10244069B1 (en) | Accelerated data storage synchronization for node fault protection in distributed storage system | |
| CN100471112C (en) | Memory-network memory-magnetic disc high speed reliable storage system and its reading/writing method | |
| Shu et al. | Th-dpms: Design and implementation of an rdma-enabled distributed persistent memory storage system | |
| EP3679478A1 (en) | Scalable storage system | |
| Barry et al. | Implementing journaling in a linux shared disk file system | |
| US8108580B1 (en) | Low latency synchronous replication using an N-way router | |
| CN101212490A (en) | Storage device controller | |
| Zhang et al. | Leveraging glocality for fast failure recovery in distributed RAM storage | |
| US20050081092A1 (en) | Logical partitioning in redundant systems | |
| US6728818B2 (en) | Dual storage adapters utilizing clustered adapters supporting fast write caches | |
| Duan et al. | Gengar: an RDMA-based distributed hybrid memory pool | |
| CN101859232A (en) | Variable length data memory interface | |
| WO2024045643A1 (en) | Data access device, method and system, data processing unit, and network interface card |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20090318 Termination date: 20210326 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |