CA2071130A1 - Communication networks - Google Patents
Communication networksInfo
- Publication number
- CA2071130A1 CA2071130A1 CA002071130A CA2071130A CA2071130A1 CA 2071130 A1 CA2071130 A1 CA 2071130A1 CA 002071130 A CA002071130 A CA 002071130A CA 2071130 A CA2071130 A CA 2071130A CA 2071130 A1 CA2071130 A1 CA 2071130A1
- Authority
- CA
- Canada
- Prior art keywords
- unit
- hybrid
- units
- link
- distribution
- 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.)
- Abandoned
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L45/00—Routing or path finding of packets in data switching networks
- H04L45/02—Topology update or discovery
- H04L45/04—Interdomain routing, e.g. hierarchical routing
Abstract
Abstract Communication Networks Fig. 1 A communication network consists of end units (EUs) and distribution units (DUs) coupled together by links which may include local area networks (LANs).
The units maintain neighbour tables by sending out Hello messages which indicate the unit type and contain the network service access point (NSAP) IDs or addres-ses of the units. Data messages (packets) mainly originate and end at EUs. An EU maintains only partial routing information about neighbors while the DUs collectively maintain complete information about all NSAPs. So if an EU wants to send a pocket to an EU which is not a neighbour (and sometimes even if it is), it need merely send it to a neighbouring DU; that DU, along with the other DUs, must find a route to the destination EU. This invention provides a hybrid unit (HU), which acts as a DU to EUs but as an EU to DUs. The HU provides DU-like message forwarding but only on a local basis within a subnetwork of EUs and HUs.
Figure 5
The units maintain neighbour tables by sending out Hello messages which indicate the unit type and contain the network service access point (NSAP) IDs or addres-ses of the units. Data messages (packets) mainly originate and end at EUs. An EU maintains only partial routing information about neighbors while the DUs collectively maintain complete information about all NSAPs. So if an EU wants to send a pocket to an EU which is not a neighbour (and sometimes even if it is), it need merely send it to a neighbouring DU; that DU, along with the other DUs, must find a route to the destination EU. This invention provides a hybrid unit (HU), which acts as a DU to EUs but as an EU to DUs. The HU provides DU-like message forwarding but only on a local basis within a subnetwork of EUs and HUs.
Figure 5
Description
20~1130 C~ o nn m ~lrli c oot 1 o ~ N e3t ~v o r~lc~
The present inventlon rel9te9 to d~8it~1 communlcation Systems consistin6 of ~ plursliSy of nodes interconnected by e communlcatlon network.
Generoi b~ck~round A communlcctlon system of the type with whlch the present invention i5 concerned conslsts of o plur~lity of unlts termed end unlts ~EUs) and dlstrlbu^
tlon units (DUs) connected to~ether to form ~ switchlng network~ Digltcl deta message~ are sent between the unlts by means of the network. (For convenience,we wlll hera use the term "packet" or "d~t~ messo~e" for messa~es convevin~ detebetween users, ~nd the term message" or "control messoge" for control messa~a3 concerned only wlth the lnternal mor.~gement of the system.) In p~rticular, w~are concerned with communlc~tlon uslng ~ network servlc~ thet ls connectlonless;thrt ls, in which each packet sent between cny pelr of unltq 19 tre~ted indepen-dbntly of sny other such p~cket by the network.~
One distinctlon between an end unlt snd ~ distrlbution unlt ls th~t on end unlt only forwards pcckets onw~rd~ lnto the network that lt itself has orl~in~-ted. A -distrlbution unit forwards both packets th~t lt orlgin~tes and p~ckats thot it receives from other unlts. Typlcslly ~n end unlt hss only one llnk connectln~ lt to the network, whll~ o distributlon unit hss many links. The end unitC and the distrlbution unlts correspond respectively to the end systems ond lntermedlete systems of sn ISO standsrd (9542), but lt will be reollzed thotthe present princlples are not llmited to syqtems conforming to that standard.
Thé operctlons of e communlcstlon system can generally be separated lnto s number of layer~ - tha stendard loyer model 1Q the 7-lsyer modbl of an ISO
standord (7498). The present prlnclple~ ~re not conflned to th~t model, but lt iQ convenlent to use lts termlnology. WQ are prlm~rily concerned here with layer 3, the network layer, which is concerned with th~ trensfer of psckets between units. Every unit has an identlfication ~t lcyer 3 whlch ls unique lnthe whole system; for convenience, wa cen USQ the ISO stsnd~rd term NSAP (net-work service ~ccess point) for this ldentificstion.
` 2~7~130 B~ckground - end unlts ond distrlbution unilts Every end unit must of course hflve en NSAP. Often an end unit' will have only one NSAP, but for o vsriety of ressons, an end unit can have a plu-rclity of NSAPs. Distributlon units will normally also have NSAPs, so thot they ccn treated os end units for purposes ~uch as system control ~nd malnten-ance (system oana~ement).
When an end unit wents to send 3 packet to another end unit, lt constructs the pscket with a header or control sectlon includin~ its own NSAP and the NSAP of the end unit it ls sendin~ the p~cket to - thst is, the-source snd des-tinatlon NSAPs. In general, the trcnsport of ~ pocket throu~h the communico-tion network between two end units involves a number of distribution units.
Each unit (end units as well as distribution unlts) maintains some routln~
informotion, conventionally in the form of routin~ tables, which enables thst unit to-decide which of its neighbouring unlts it should forw~rd the packet to as the next hop in its path tG the ultimote destination. The routin~ t~bles ~ssoci~te each possible destination NSAP with some forw~rding information. Thelatter includes the identificstion of the link over which a packet containin~
particular destination NSAP should be forwarded.
If the link has multi-occess cspability ~such as 9 Locsl Area Network (LAN) link which permits many units to connect to the same link), then the forwarding informotion also inrludes the llAN identifier of the particular nei~hbourins unit on thet l~nk to which the packet should be forworded. (This LAN identifier is o layer 2 oddress, an~ is lndependent of the layer 3 ~ddresses, NSAPs, with which we ere primarily concerned.) In genersl, the routin~ tobles of the distribution units must maintain complete routin~ information obout ~11 units in the network. This is essentialfor the core network to be oble to pass (dot~) packets to all end units; if the core network does not have a record of the existence of an NSAP, it obviously csnnot locate it (sddress it) and so cannot p85S p9ckets to lt. The routin~
tables of the end units need not, howeYer. meintain complete lists of all the distribution units to which they ~re connected. It is desiroble for them to hold a re~sonable number of distribution unit addresses, but ~ sin~le distribution 20711~
unit address would be sufficient to 9110W the end unit access to the core net-work.
When an end unit wents to send a p~cket, lt looks in its routing tables for an entry with the destination NSAP. If lt finds uch an entry, it forwsrdQ thepocket usin~ the forwarding informotion ossociated with th~t entry ln the t~bles.
If it does not find a match, it chooses one of the distribution units present inits tables and forw~rds the packet to thet distribution unit. At the qsme tlmelt makes sn entry in lts routin$ tables for th~t NSAP with forwardln~ lnformo-tlon set to that of the chosen distribution unlt.
If the receivln~ distribution unit finds from its routin~ tables that the unit to which it should forw~rd the packet is on the same link ~s the packet was received on. then it may send a speciel redirect control mess~e to the source end unit, informin~ it that it c~n forward direct to the next hop unit for packets sent to th~t NSAP. The end unit receivin~ the redirect control messagethen upd~tes its routing tables entry for th~t NSAP with the forwardin~ infor-mation provided by the distribution unit.
A distribution unlt contalns o considerable amount of routing information in its routing tables. In principle,-an end unit need only contain the identi-fier of a sin~le distribution unit in lts routin~ tables. However, it is conven-lent for it to contain some routin~ information. This information will controlthe selection amon~ the distribution units in dependence on the destinstion NSAPs or, if the destin~tion unit is ~ neighbour of the end unit, it will contain the forwerdin~ informstlon necessory to send directly to the destinstion unlt wlthout pessin~ throu~h ~ distributlon unlt. Thus the routes to destinationc can be optlmized.
B~ck~round - rou~ng tuble m~intenance ond Hello mçss~ges There is a vsriety of ways in which the routin$ tables can be $enerated and maintained, The routin~ functions of end units are concerned only with the lniti~l hop of the path of a p~cket - from that end unit, to either the des-tinstion end unit or some distribution unit. The routing functions of distri-bution units c~n be divided into two cl~sses: those concerned with the finol hopln the pDth of s p~cket (from that distributlon unit to the destin~tlon end unit), 207~13~
~ 4 ond those concerned with intermedi~te hops (DU-DU hops, to further distribution unlts).
To deal first with DU~DU hops, there 9re m~ny ways in which routing infor-matlon for these hops m~y be m~int5ined ln distribution units. These units typically exch~nge control messages, formin~ a routin~ protocol, which contain routin~ information, and esch psrtlcipstes in a distributed routing al~orithmwhich ensures that eoch distribution unit will construct routin~ t~bles thst arecompatible with those in each other distribution unit. The compstibillty ensures that, for example, p~ckets will eventually be delivered to their destin~-tion ond will not repetitively circulQte around a closed loop of distribution units.
End units typically do not particip~te in the routin~ algorithm used by the distribution units, relying instead on their neighbouring distribution units to forw~rd packets appropriately. Thus end units typic~lly have much smoller and simpler routin~ tables thon distribution units.
We ~re primarlly concerned here with routin~ involving end unlts - EU-EU, EU-DU, and DU-EU hops. For this, the routin~ functions requlre knowled~e of the existence of neighbouring units. Such information may be confi~ured into e~ch unit by sn operator or lt may be learned dynamicslly through exchan~e of messa~es. The present system is concerned with networks in which this infor-mation is learned dynomicslly throu~h the exchange of control messages c~lled Hello messoges. Such messo~es are sent to neighbouring units but cre never forw~rded onwards through the network.
An end unit sends out EU-Hello messages to announce its presence to neighbourin3 distributlon units; ~nd similerly, a distribution unit s~nds out DV~
Hello messa~es to announce its presence to neighbourin~ end units. Option~lly,units of g psrticuler type cdn also note the presence of nei~hbourin~ units of the same type (e.g. end units can note the existence of other end units by listening to EU-Hello messsges~.
The Hello messa~es sre sent out st regulsr intervals, to confirm the con-tinuing presence of the sendin~ units. Ecch receiving unit contsins timers forthe various sending units which it is receivin~ Hello messoges from, to check that it continues to receive the ~ppropriate Hello messa~es; lf a timer tlmes out, 2071 13~
: - 5 -then the receiving unit deletes the correspDnding sendin~ unit from Its records.Esch unit is in ~ener91, of course, both a sendin~ unit and a receiving unit for Ha~ ~OSS~QS, The timer rste is norm~lly set to be lower thDn the rate ot which the associated type of sending unit sends Hello messc~es, so that on occ~sion31 lostHello message does not result ln the receiving unit deleting o sendin~ unit fromits records. If c distribution unit should dlsoppeor, that foct needs to become known to other units quickly, to ovoid packets being forwarded by mony unlts to nei~hbouring distribution unit thot hos stopped working; if on end unit should cease workin~, only p~ckets destined for that end unit will be lost. Distribu-tion units therefore normally send out Hello messsges much more frequently thsn end units.
Each EU-Hello mess~e includes the NSAPs of the end unit produclng it.
Any unit which receives an EU-Hello message records the NSAPs ln its routin~
tcbles. It also records, ogoinst each such entry in the tsbles, the forwordin$
information necesssry to send ~ packet to thst end unit. Norm~lly, the forw~r-din8 informotion includes ot least the identification of the link over which theEU-Hello message w~s received. If thot link ls o multi-occess link, then the l~yer 2 addressin~ info~rmotion necesssry to send a packet over that link to th~t end unit snd derived from the received packet m~y also be included in the for-wsrdin~ information, olthou~h it is possible thât this informstion could be derived subsequently by some other me~ns.
Each DU-Hello messoge will normslly include the NSAPs of the distribution unit producin~ it (though in some vorionts of this scheme, such informotion is not necess~ry and so is not included). Each end unit which receives these messoges records them in its routin~ t3bles, to~ether with the ssme forwording informâtion described obove.
In summ ry, then, the present type of communicstion system essentially involves two types of unit, the end units and the distribution units. The end units contsin routin~ tsbles which cont~in ~ rel~tively small ond possibly incomplete ~mount of lsyer 3 informotion re~ordin~ immediately cd~scent units;
the distribution units contoin routin~ t~bles which collectlvely contoin complete informotion about ~11 the units in the system ond routes between them.
6 2071~3~
The present invention The crux of the present invention i~ the provi~ion of a further type of unit which can be connected between end units and a distribution unit and which appears as a di~tribution unit to the end units and as an end unit to the diqtribution unit. We will call this new type of unit a hybrid unit (HU). A hybrid unit ha~
only one link to a di tribution unit, but can havs a plurzlity of links to which end units are connected (directly or via .further hybrid units).
The invention in its broad form i~ a3 follows:
In a communication network comprising end units (EU ) and distribution units (DUq) coupled together by communication links, each end unit having at lea~t one network service acces~ point (NSAP) or address which identi~ie~ the unit and each distribution unit including routing tables for routing data packets through the network between end units, the unit~ maintaining knowledge of their neighbouring unit~ by exchanging Hello messages which indicate the unit type and contain the NSAP (9) of tha end units, the DUY
collectively maintaining complete information about all NSAPq, a hybrid unit (HU) having one link (HUE link) on which it simulate~
an EU and a plurality of links (HUD links~ on which it ~imulate~
DUs, ~ending on the HUE link ~U-Hello message~ containing the NSAP~
of all EU~ which it receives on its HUD link~, and sending on the HUD links DU-Hello messages.
The links of a hybrid unit to which end units may be connected appear a~ links to a distribution unit to those end unit~; the link 6A 20711~0 by which the hybrid unit i~ connected to a distribution unit appears a~ the link to an end unit to that di~tribution unit. Thi3 combination of features means that the hybrid unit can be used in a communication ~ystem of the present type without requiring any modification of the existing units of the ~y~tem.
A~ described in d~tail hereinafter, a hybrid unit combines certain properties o~ end units and of diqtribution unit~. The hybrid unit provides some o the function~ of a distribution unit, but a major distinction b~tween the hybrid unit and a di~tribution unit is that the hybrid unit contain3 only a limited amount of routing information compared to a distribution unit.
The routing information of the hybrid unit i8 derived only from information obtained from its neighbourQ, and is similar in qcope to ~hat maintained by an end unit. In contrast, the routing information of a di~tribution unit contain~ all po sibls destinations in the network. The routing information processing which the hybrid unit ha~ to perform i~ thexefore much simpler than that which a distribution unit has to perform. The complexity o a hybrid unit is therefore closer to that of an end unit than that of a distribution unit.
The hybrid unit as 80 far de cribed may be of fixed configuration. It can however be made reconfigurable, either manually or automatically. Such a reconfigurable hybrid unit further enhances the flexibility of communication systems in which it is used, particularly when the syqtem changes (a3 the re~ult of either deliberate changes to its topology or failureq of part~ of the system).
6B 2 0 7 ~ 1 3 0 Since di~tribution unit~ are complex, it i~ desirable to minimize their number. If thers i8 a modest number of end unitR in an area which is rea~on-20711 3~
.
ably compect ~eogrsphicslly, thls c~n be achieved by coupling the end unitsdirectly to a single distribution unit. If the number of Pnd units in such an sree is l~rse, this technique becomes inconvenient. is the number of ~orts available on a distribution unit is generally fairly smsll. However, a 13r~se number of end units can be connected to a sin~le distribution unit through a multi-access link such as a LAN, as descrlb~ obove.
If the end units are widely dlstributed $eogrophically, however, then it is $enerally not feasible to connect them to a single distribution unit elther directly or by a LAN. With o conventlonol system, they must be grouped into ~eographicall~J small groups, each of which has a sepsrote distributlon unit.
The hybrid unit effectively provides an altern~tive w~y of connectin~; end units in such a situatior.. Esch geo~raphically smsll group of end unlts is connected to a hybrid unit, and the hybrid units are in turn connected to o sin~le distribution unit.
BPIE~ D~ IO~I OF D~L7~
Two types of hybrid unit, a simple (non-outoreconfigurable) unit and an autoreconfigurable unit, embodying the invention will now be described, by way of example, with reference to the drawin~;s, in which:
Fi8. 1 is a block dia8ram of a communication system includin~ hybrid units;
Fi8. 2 is a block dia~ram of an end unit, wlth particular reference to its routing tsbles;
. .
Fi$. 3 i4 a block dia8ram of ~ distrlbution unit, with particulcr reference to its routing tables;
Fig. 4 ls e block diegram of a simple hybrid unit, with psrticular reference to its routing tablesi Fig. 5 is 9 block dis6rsm of the components of an Dutoreconfi~urable hybrid unitadditional to those shown in Fi~. 4;
Fig. ô is a $eneral flow dia~ram of the operotion of the hybrid unit of Fi~;. 5;and ` 2071130 Fi~s. 6A to 6C show the blocks of Fi~. 6 in more detail.
General system Fi~. 1 shows a communication system includin~S four distribution units DU1 to DU4. Distrlbution unit DU1 is connected to the LAN link LANI, which ~lso has three end units EU1 to EU3 connected to it. Distribution unit DU2 is con-nected to a further end unit EU4 ond olso, through seporote links, to the LAN
link LAN1 and o further LAN llnk LAN2. Distribution unit DU3 is connected to three further end units EU4 to EU7; the end unlt EU4 i5 also connected to dls-tribution unit DU2 (as just noted).
The system as described so fsr is conventional, ond shows o variety of features of convention~l systsms (slthou~h many systems do not include all th sevarious feotures).
End unit , Fi8. 2 is a block dio~rom of the routing tables and ossociated components of an end unit (say EU1). The main components are a local NSAP list store 10;
a control messa~e receiver unit 16, whose purpose is receiving control messa~es;two routin~; tables, e distribution unlt table 19 and an NSAP table 20; end a route look-up en~ine 22.
:
The local NSAP list store 10 contoins the NSAP(s) of the unit, snd is coupled to an EU-Hello messa$e ~enerotor 11, which is olso fed from on EU-Hello timer 12. At sultoble intervals determined by the timer 12, the messa~e gener-ator 11 assembles EU-Hello messages, containin~ the NSAP(s) of the end unit, andtransmits them on the links from the end unit (e.~. a LAN, or links such as thatfrom end unit EU5 to distribution unit DU3 or those from end unit EU4 to dis-tribution units DU2 and DU3).
Eoch location in ths distribution unit table 19 consists of two sections, a timer section 19-T and a distributior, unit forwardin~ information section 19-E.Any incomin~ DU-Hello message from a dlstribution unit neighbourin~ the end unit is received and proressed by the receiver unit 16. This unit writes, in section l~F of table 19, in~ormation obout the exiQtence of the distribution unlt, together with the ~orwarding informotlon neeess~ry to send o pocket to the 207~130 g dlstribution unit. As a result, the tabie 1 g will gredu~lly accumulate the identifiers of the distribution units coupled to the end unit. The number of entries in the tabla may be restrlcted so that only the most recentlv hesrd'fromdistribution units are contcined in the table, for example.
The unit 16 also writes into the timer section I 9-T of each locction In the distribution unit table 19 the time tfrom a deletion clock unik 17) when the last Hello message for that location was recelved. The deletlon clock unlt 17 constantly comp~res the stored times ln the tlmer section 19-T with the current time, and erases the entry from the dlstribution unit table lf the time difference exceeds a preset value. (This value moy be derived from g field in the DU-Hello, so that it can be different for different distrlbution units in the table.) Thus the contents of the distrlbution unlt table are maintained to match the active distribution units currently neighbourin~S the end unit.
I~ch location in the NSAP table 20 consists of three sectlons, a timer section 20-T, an NSAP section 20-ID, snd s forwarding informstion section 20-F.
For each NSAP contained in the table, section 20-F contains forw~rdin~ informa-tion for use when sendin~ packets destined to that NSAP. A deletlon clock unit 23 operates, in conjunction with the timer section 20-T of each location inthe NSAP table and the wrltin~ lnto those sectlons of the last update times, to delete obsolete entries from the tsble.
When the end unit wants to send a data packet, it f irst uses the route look-up en~ine 22 to look for an entry in the NSAP t~ble 20 whose NSAP (in sectlon-: 20-ID) matches the destin~tion NSAP for the pecket. If such an entry exists in the NSAP table, the end unit uses the associsted forwording inform~tion in section 20-F to send the packet to the next hop neighbourinE~ unit.
If there is no entry in the NSAP table for the destination NSAP, the route look-up en~ine selects an entry, e.~. at random, in the distribution unit table 19.
It adds a new entry to the NSAP table 20 with the NSAP section 20-ID set to the destination NSAP of the packet, the forwardin~ sectlon 20-F set to the for-wardin~ information extracted from the selected information in the di,tribution unit table 19, and the timer section 20-T set to a suitable value.
If a distribution unit respond~ by sendin~ a dlstribution unit redirect messo~e to the ori~;inatin~; end unit, this ls passed by the control messa~se 2 ~ 3 ~
~o -receiver unit 16 to a distribution unit reclirect en~ine 21, which changes the forwarding inform3tion in section 20-F of the NSAP table 20 to that specified inthe dietribl tion unit redirect message and resets the ~ssociated timer in section 20-T (to 9 preset v31ue). When the end unit st~rts up, the routing tsble 20 is initially empty. As packets are sent out to different destinations, so entries graduslly accumulate in this table.
E~ch time an entry is mode in the NSAP table 20, the timer section 20-T
for that entry is updsted. Entries in this t~ble eventu311y time out and are deleted by the deletion clock 23. They are then reformed ss described above when new packets are sent by the end unit.
The size of the NSAP table 20 is usually limited. If elther the route look-up en~ine or ths distribution unit redirect engine needs tn make a new entry in this tsble and it ls full, then an entry is deleted; The entry for deletion may be chosen at rondom or based on the velue of its ~ssocieted timer section 20-T.
As noted sbove, it is possible for end units to listen to EU-Hellos. This provides another source of information for msintaining the NSAP table 20.
This will, however, involve fairly substantial processing overheads, and is liable to fill up the table with NSAPs which the end unit is unlikely to want to send psckets to, so in practice this is not usually done.
Distribution unit Fi8. 3 i9 0 block diagr~m of the rout1ng tables end associoted components of a distribution unit (say DU1). The distribution unit has a local N~AP table 31, a remote NSAP table 37, a control message receiver 32 (which is mainly for EU-Hello messages), ond a control messa~e generator 33 (which is for DU-Hello messages and redirect messsges).
The DU-Hello messages are genersted in unit 33. These are transmitted on all links of the distribution unit at suitable intervals, as determined by a timer 34.
The local NSAP table 31 contains ~n NSAP section 31-ID, a timer section 31-T, ond o forwardin~s inform~tion section 31-F. Thls table contain~ the NSAPs _~ 1 ~ :~ Q ~
_~
_ _ _ ~ I
of er"i urlits wr:~-h are nei~hbours ot` the distribution unit~ ElJ-Hello messâges 3re rP eive.i b; :.r,it - ar,~ P ~ISAP¢s) arl-l as oci3ted forwarding ir,formatior, ~' are written into the appropriate ecti:)ns :f the tJ~le ~1l alon~ with the tlnle :;
receipt. Thus he table ~11 will gradu311y accumulate thê identifiers c.f the encl ~F units :oupled t.-. ~ ;-,e distribution unit. A d-letion clock unit ~c; rr,aintains r these entrie .urrent in the samP W2 ~ as the deletion clock unit 1, c f the end unit shown in FiF. '.
The distribution unit also contâins a remote NSAP table 37 which like table ~1 contain entries associatin~s NSAPs (in ,ection ~7-ID) with forwarding inform3tion (in ,e.tion 3,-F~. Entries in this table are made and deletêd by a routing control -ngine ~ which implements the distributed routing algorithm used by all the distribution units. This table contains the NSAPs of end units which are not neighbours of the distribution unit. ~In practice. the orgâniza-tion of this table may e;~ploit the generâlly hierarchical nature of NSAPs.!
When the distribution unit receives a packet it matches the destinetion NSAP of the packet against the entries in sections 31-ID and 37-ID of the routin~s tables using the route look-up en8ine 3fi. The forwarding informatior.
is obtâined from the appropriate routing table. and the packet i5 forwarded towards the destination. If the de-tination ~ISAP is in the local table 31. the destination end unit is 3 neighbour of the distribution unit. If it is in the remote table 3, then the end unit is not a nei~shbour of the distribution unit so the forwarding information will cont3in details of the next distribution unitto whi h the pack-t is to be forwardeà.
If the forwarding information used by the distribution unit results in the pa-ket being forwarded to another unit connected to the same link as that on whic.' tr.e p~ l_r WâS r cei~ d r,- -u;.irg 'ol~-a? -n~ir,e ;f~ . ner3t-s a reiire:t rr)ess2g -oritdir.irl;r th- destir.3ti~rl NSAF 3n] ~he f~.~ wardir,~- inform3-: n n,atcr,ir,g that `;SAF in the routin. ~Dle_. ar,d sends the rr,es iage t; tne 5~ ir~e end unit.
The operation of the systern as described 50 far is onvêntional 3nd shows a variety of features of conventional systems ~although many systems do not include all theese various features).
2~711 30 Hybrid unitq - ~eneral Returnin~; t,~ Fi~. 1. the syst~m a1so includ~s ~ distribution unit r~U4, wi+h a nei~hbourin~ end unit EU7 snd ~lso two nei~hbouring hybrid units HU1 snd HU2. Hybrid unit HU1 has fl nei6hbourinE~ end unit EU8 ~nd ls also attached t LAN link LAN2. Hybrid unit HU2 has a neighbouring end unit EU11 and a further nei~hbouring hybrid unit HU3. Hybrid unit HU3 has two nei$hbouring end units EUl2 and EU13. These parts of Fi~. 1 show a variety of ways in which hybrid units may be used (~lthou~h mony systems will not include all these vorious features).
The hybrid units are strictly hierorchlcal, with a sin~le superlor link snd a plurality of subordinates allowed. On its superior link, ~ hybrid unit appeq~C 95 an end uni+; thus hybrid units Hlll and HU2 sppeor 9S end units to distribution unit DU4, sending EU-Hello messages up to it and receivin~ DU-Hellomesss~es from it. To its subordinates, o hybrid unit appeors ~s a distribution unit; thus hybrid units HU1 to HU3 oppeor as distribution urits to end unlts EU8 to EU13, sendin~ out DU-Hello messa~es to those end units and recelving EU-Hello mess~ges from them.
In principle, HU1 could have more than one port appearin~ as an end unit (rather like end unit EU4>. This would result in multiple paths from the end units subordinate the hybrid unit. In the csse of a hybrid unit, this would produce no significant sdvanta8e and would result in various complications, so we prefer not to implement this possibility.
, .
Hybrid units can be coscAded, as shown by hybrid units HU2 snd HU3. The same hierarchic rules apply, so that hybrid unit HU2 appears as ~ distribution unit to hybrid unit HU3 and hybrid unit HU3 appeers as ~n end unit to hybrid unit HU2, with EU-Hello messages passing upwords from HU3 to HU2 ~nd DU-Hello messa~es p~ssing from HU2 to HU3.
A hybrid unit performs both Hello messa~e and dats packet routing func-tions. Considerin~ first its response to data packets, it forwards these in bothdirections. It performs a routing function on these, forwardin~ packets coming from the distribution un~t above it on to the ~pproprlote link to the destination unit, and forwording packets comin~ from the end units below it either on up to the distribution unit above it or directly out downwords to the destinatlon unit.
2~`13~
Considerin~ its routln~ control functions, the hybrid unit collects the NSAPs of its subordinate end units received throu~h EU-Hello mess~ges ~inclu-din~ thocr whirh are- indirectlv subordinate through further hvbrid units). anclappears as a sin~le ordinsry end unlt ~albelt with sn unusually large number of NSAPs) to its superior unit (whether a further hybrld unit or a distrlbution unit) by psssing on, in EU-Hello messages whlch lt sends to it3 superior, all the NSAPs it hes collected in this way.
Simple hybrid unlt FiS. 4 is a block dlegram of the routing table and assoclated components of a simple (i.e. non-3utoreconfl$urable) hybrid unit (say HU1). This contalns three routing tables: a distribution unit table 45 and an NSAP table 54, corres-pondinP resp.oçtivel~f to the distribution unit table 19 and NSAP table 20 of the end unit of Fi~. 2, snd an end unlt table 49, correspondin~ to the local routingtable 21 of the distribution unit of Flg. 3.
Considering first the upward flow of messages, an EU-Hello message receiver 53 receives EU-Hello messa~Ses from end units and posslble hybrid units lower inthe hierarchy, and feeds these messa~es lnto the end unlt table 49. This table hes three sections, section 49-ID for the NSAPs contained in the incoming mes-sages, section 49-F for nssociated forwarding information, and section 49-T for the times of receipt. A deletion clock and timer unit 52 provides these times and deletes entries in the table which are not updated by fresh EU-Hello mes-sages at suitable intervalQ. The teble 49 must be lar~e enough to retain all NSAPs-of 911 the end units below the hybrid unlt. (This is like the routing tables of the distribution units, and in contrast to the routin~ tables of the end units.) An EU-Hello generator 42 i~ coupled to the table 49 and to 9 local NSAP
list store 43 which stores the NSAP(s) of the hybrid unit itself. At intervals determined by a timer 41, this generates EU-Hello messa~es which include all theNSAPs in end unit table 49 and those in store 43 and transmits these to the distribution unit or superior hybrid unit.
Considering now the upward flow of data paclcets, if the hybrid unit receives a data packet coming up to it from an end unlt or subordlnate hybrid unlt, the route look-up engine 50 inspects the header of the packet to determine ~7~39 whether there is 9 match between its destinotion NSAP and any N5AP in the end unit tsble 49. If there is, then the hybrid unit forwRrds the packet on the appropri3ta output link using the forw~rdins information assocista:l with th~t NSAP in end unit table 49.
If there is no match in the end unit table 49, the route look-up en~ine 50 looks for a match between the destination NSAP of the p~cket and any NSAP in the NSAP table 54. If there is, then the hybrid unit forwards the packet on the sppropriate output link using the forwarding information associated with that NSAP in the NSAP table 54. T~ble 54 ls updated by redirect messa~es received by a redirect messa~Se receiver 51 from a distribution unlt or superiorhybrid unlt and by a deletion clock unit 55.
Tf there is no mstch in either of these tables, then the hybrid unit for-wards the packet up to a distrlbution unit or superior hybrid unit using the some algorithm as that used by the end unit. The route look-up engine 50 enters the destinstion NSAP into the NSAP tsble 54 to~ether with the forwardin~Sinformation associated with the distrlbution unlt.
The route look-up englne 50 may ~lso ~enerate updating redirect control messages like those 8enerated by the distribution units, as discussed above, sendlng these to the end units and subordinate hybrid units.
Considering now the downward flow of data packets, the hybrid unit inspects the header of a pscket received from a distribution unit or superior hybrid unit, matches~.. the NSAP against the NSAPs ln the end unit table 49, and forwards thepacket downwards according to the forwardin~ inform0tion associated with th NSAP in table 49.
Considering finally the downward flow of control mess~es, the hybrid unit receives DU-Hello messages sent by superior distrib~Jtion units or hybrid units and these ere processed by the DU-Hello message receiver 44 and stored in the distribution unit table 45 (in exactly the same way as an end unit processes these messages). On those links to subordinate units (hybrid unit;, or end units), the hybrid unit periodically sends DU-Hello messa~es using the DU-Hello gener~tor 47 controlled by the DU-Hello timer 48. .
2~7~
The hybrid unlt also receives distribution unlt redirect messages through the distribution unit redirect messa~e receiver 51 and modifies the information in the N!~AP table F~4 based on the NSAP 3nd fc rwardin~ information received in the distribution unit redirect messa~e.
The hybrid unit thus appear as on ordinary distribution unit to the end units ond ~ny subordinate hybrid units, and os an ordinory end unit to the superior distributlon units or superior hybrld units.
It will be noted that if there should be a foilure of the superior unit or units of a hybrid unit (i.e. of the distribution or superior hybrid unit or units), the port of the network includinES ond below the hybrid unit can continue to operate. The hybrid unit con continue to forword pockets between the end units below it.
In ~eneral, each hybrid unit has severol links. One of these links is that connected upwords in the hierorchy, to o distribution or a superior hybrid unit, and behavin~ as on end unit link, and we can term this an HUE link. The remoinlng links oll behove os links of e distribution unit, ond we con term themHUD llnks.
Autoreconfi~5uration We have assumed so far that the HUE or HUD noture of the various links of the hybrid units is predetermined. In proctice, it is convenient for the nature of the links not to be predetermined. This allows the hybrid unit to have its links connected arbitrorily. If this is the case, then the hybrid unit must be confi~ured so thot the appropriate link becomes its HUE link and the remainin~ links become HUD links.
The simplest form of configuroble hybrid unit is one in which the configu-ration is operotor determined. This may require dlrect manual settin~ of the links, or it may be performed by control messa~es ~eneroted by the operator but sent throu~h the network to the hybrid unit.
In large ond complicoted systems, there may be more thon one possible woy to configure some of the hybrid units, or it m&y be desirable or necessary to change their confi~urations in the event of a network change (which may be a 2~7~ 3~
~ 16 --permanent addition to the network or deletion of some part of the network, or may be a temporflry change resultin8 from a temporary failure or recovery from a failure, fc~r example of a link~. s If, in the Fi~S. 1 system, distribution unit DU4 were to fail, end unit EU8 would become in~ccessible. It would therefore be desirable to reconfi$ure - hybrid unit HUl to presert an HUE link to the LAN link LAN2, and hence to distribution unit DU2, so m~king end unlt EU8 occessible agoin. Such reconfi3-urotion can be achleved by operotor control, ~ust ~s with initial confi~uration.
It will be reclized that the fcilure of distributlon unlt DU4 makes end units EU7 and EUl 1 to EUl 3 inaccessible to the rest of the system, ond this connot be rectified by ~ny reconfi~uratlon. End units EUl 1 to EU13 remoln in c4mmunication with each other, however, Vi3 the hybrld units HU2 and HU3.
The Fig. 1 system could be configured with hybrid unit HUl presenting an HU llnk to distribution unit DU2 via LAN link LAN2, and bein~; effectively dis-connected from distribution unit DU4 even if that unit were fully functional.
This could result, for exsmple, if dlstribution unlt DU4 recovered after its failure.
Direct manual reconfi~uration is often f~r from convenlent, because the part of the system where 3 failure occurs may be remote from an operator cap-able of performin~ the reconfi~Suration. Reconfiguration by remote operetor control is also often not satisfactory. It may be difficult or even impossible for a remc,te operator to determine the precise nature of the system failure; ~nd bec3use the system hss under~one o feilure, lt msy be difficult or impossible ~chleve the optimum reconfiguration remotely.
The hybrid units therefore preferably include autoconfiguration meflns, whereby they confi~ure the HUD or HUE nature of each of their links automatl-cally.
If a hybrid unit has no hybrid unit as neighbour, autoconfiguration requires only that it should find a distribution unit to which it can present ~nHUE link. Thus hybrid unit HUl con autom~tlc~lly rsconfi~5ure itself, in the event of ~ failure of distrlbution unit DU4 or its link to that unit, simply by 2071~
~ 17 --makin~ lts link to LAN link LAN2 an HUE link, so becoming subordinate to dis-tribution unit DU3.
If a hybrid unit has a hybrid unit neighbour, however, then it msy be desircble to reverse the HUD-HUE direction of the link between the two hybr$d units; and if both the hybrid units have distribution units a5 neighbours, then this mey be necessery. Further, the possibility of a circulating peth for data packets must be ~voided. (This could hsppen if, for exemple, there are three hybrld units ell connected together.) In generel, therefore, en autoreconfi~uro-tion algorithm is required which results in the system remaining connected as far as possible while cvoidin~ inconsistency.
There are o; course various possible 13eneral algorithms for autoconfiguring a network containin~ hybrid units. The ma~or ob~ects to be achieved in genercl can be defined in various ways, such as:
there should be no closed loop of hybrid units with HUD to HUE connec-tions;
2 no hybrid unit should have more than one HUE link; and 3 a~ many hybrid units as possible should have HUE links.
If connection to the network (i.e. to a distribution unit) is possible, then condition 3 means that every hybrid unit has en HUE link, and condition 1 then means that all hierarchically upwerd paths through hybrid units must eventually terminete at a distribution unit. If connection to the switching network is not possible, then these conditlons result in precisely one hybrid unit not having an HUE link, i.e. becoming the master hybrid unit, wlth all the other hybrid unit forming hiererchically upward paths to that mester hybrid unit.
We will describe a mechsnism using c spanning tree algorithm which permits erbitrary topolo6ies of distribution units, hybrid units, and end units. Simpli-fications of this mechanism are possible if the topology of the subnetwork of neighbouring hybrid units, plus the distribution units end end units neighbour-ing the subnetwork, is restricted such that there are no alternative paths between any two units, and there is only one link from ~ connected set of hybrid units to a distribution unit.
2~7~130 ~ 18 --To permlt autoconfi~uration, a new control message is defined, termed e HU-Hello. This allows neighbourin8 hybrid units to reco~nize each other as hybrid units, snd rc~ntains some inform3tion wh~ch aLlows hybrid units tc) USa 3 spansin~
tree algorithm. Hybrid units use the algorithm by combining information from the HU-Hellos they receive with their own local information.
The spanning tree algorithm ~enerates a singly-connected logical topology for the subnetwork from the srbitrQry physlcal topology, ond in sddltion forces each hybrid unit to a8ree with its hybrid unit neighbours the HUD/HUE statss of its links so that the hybrid units of the subnetwork act together correctly.
The algorithm selects e single hybrid unit as the root of a tree spanning the subnetwork, choosing for this purpose the hybrid unit with highest priority (as described below).
The priority scheme is desi~Sned so that the alE~orithm will operate in all hybrid unit ~ubnetworks, whether or not they hove distribution unlt nei~hbours.
If o hybrid unit has a distribution unit- neighbour, its inherent priority is replaced by a boosted priority which is hi8her than the inherent priority of anyhybrid unit. In the presence of distribution unit nei~Shbours, therefore, the e40rithm always selects as the root hybrid unit a (single) hybrid unit which hasa distribution unit nei~hbour.
The spanning tree Qlgorithm operates over the whole of the subnetwork of hybrid units; processin~ is corried out in the individual hybrid units, with communication between them for the purposes of the al~orithm bein8 by means of the HU-Hello messages. The operation of the ~lgorithm is thus distributed amon~ the various hybrid units of the subnetwork. The processin~ in each hybrid unit conslsts of a sequence (cycle~ of operations which is repeated each time the state of any nei~hbour of the unit chan~ses in a significant wey. Once a cycle has been initiated in a unit, that cycle will automatically proceed to completion. The eventual result is that the hybrid unit with the highest priority is selected as the root, to which all other hybrid units become inferior.
with a simply connected connection pattern. This may involve makin8 certein links effectively inoperative for the transfer of data messa~es.
The operation of the al~Sorithm i5 initiated whenever any change in the st~te of the system - e.~ a link or ~ distribution unit or a hybrid unit becoming operat1ve or inoper~tive - occurs. This may be detected by e hybrid 2~71 13~
,9 unit noticin~ some change - e.~. a link becoming operative or becoming inoper~-tive.
Esch cycle of the elgorithm in a unit consists of three successive sta~es, I
to III, shown as blocks 70 to 72 ln Fi$. 6. In st~e I, a priority, a root cost, and e designation are determined for each link. In stoge II, the hlghest priority link is determined~ In stage III, the state of each link is set to HUD
or HUE from the informetlon calculated in sts~es I ard II. The three sts~eC
ore performed in successlon, and the unit then enters a woitin~ stote, block 73,ond remains in thet state until o chsnge occurs. The chonge may -be ony of thestote variobles used os input to the spannin~ tree olgorithm such ns the opera-tional state of a link or the contents of o neighbour's HU-Hello messoge.
Hybrid uni~s send HU-Hello messs~es on all their links~ Nei~hbourin~ d1s-tribution units and end units l~nore these mess~ges - only neighbouring hybrid units receive and oct upon their contents. The contents of on HU-Hello mes-so~e ore link-dependent, Qnd include 9 priority field and a cost field. The priority is set to the highest priorlty from all link priorities ~excluding the link over which the HU-Hello message is to be sent~ ond the priority of the hybrid unit itself. The cost field contains the accumulated cost of the path from a hybrid unit to the root hybrid unit; this is the sum of the costs of eochof the links on the path from the hybrid unit to the root hybrid unit.
The spanning tree algorithm requires various choices to be mode between different çntities of the same kind (such as hybrid units, or links). For these choicesm eoch of the entities hoQ a value termed 9 priorlty, and the choice is m~de by selecting the entlty wlth the hi~hest priority. For convenlence, theseprlorities ore defined as unsi~ned inte~er numbers with lower integer values having higher priority. Also for convenience, seporote priorities ~re combined together in various ways to perform priority comporisons.
In perticulor, the priority field of an HU-Hello message is a concatenation of three subfields which are, in descendin~ order of si~nificance: DUbit, ~ sin~le bit which is O if the hybrid unit has a nei~hbouring distribution unit ~nd 1 otherwise; e hybrid unit priority value which is configureble by the operator;
and a unique identity value HU-ID. The highest prlority is token os the lowestvdue (treating the combined bit field~ as unsigned number4~. We will use the 2~1130 not~tion <x,y,z> for ~ prlority formed by concotenating the fields x, y, and z, with x bein~ the most significant field~
The hybrid unit priority value can be used to manually configure 9 desired hybrid unit to become the rsot in the absence of ~ny distribution unlts attachedto the subnetwork; the HU-ID field must be unique for eoch hybrid unit in the subnetwork.
Autoreconfigureble hybrld unit . .
Fi~S. 5 is a block dia8ram of the circuitry (additionel to that shown in Fi~.
The present inventlon rel9te9 to d~8it~1 communlcation Systems consistin6 of ~ plursliSy of nodes interconnected by e communlcatlon network.
Generoi b~ck~round A communlcctlon system of the type with whlch the present invention i5 concerned conslsts of o plur~lity of unlts termed end unlts ~EUs) and dlstrlbu^
tlon units (DUs) connected to~ether to form ~ switchlng network~ Digltcl deta message~ are sent between the unlts by means of the network. (For convenience,we wlll hera use the term "packet" or "d~t~ messo~e" for messa~es convevin~ detebetween users, ~nd the term message" or "control messoge" for control messa~a3 concerned only wlth the lnternal mor.~gement of the system.) In p~rticular, w~are concerned with communlc~tlon uslng ~ network servlc~ thet ls connectlonless;thrt ls, in which each packet sent between cny pelr of unltq 19 tre~ted indepen-dbntly of sny other such p~cket by the network.~
One distinctlon between an end unlt snd ~ distrlbution unlt ls th~t on end unlt only forwards pcckets onw~rd~ lnto the network that lt itself has orl~in~-ted. A -distrlbution unit forwards both packets th~t lt orlgin~tes and p~ckats thot it receives from other unlts. Typlcslly ~n end unlt hss only one llnk connectln~ lt to the network, whll~ o distributlon unit hss many links. The end unitC and the distrlbution unlts correspond respectively to the end systems ond lntermedlete systems of sn ISO standsrd (9542), but lt will be reollzed thotthe present princlples are not llmited to syqtems conforming to that standard.
Thé operctlons of e communlcstlon system can generally be separated lnto s number of layer~ - tha stendard loyer model 1Q the 7-lsyer modbl of an ISO
standord (7498). The present prlnclple~ ~re not conflned to th~t model, but lt iQ convenlent to use lts termlnology. WQ are prlm~rily concerned here with layer 3, the network layer, which is concerned with th~ trensfer of psckets between units. Every unit has an identlfication ~t lcyer 3 whlch ls unique lnthe whole system; for convenience, wa cen USQ the ISO stsnd~rd term NSAP (net-work service ~ccess point) for this ldentificstion.
` 2~7~130 B~ckground - end unlts ond distrlbution unilts Every end unit must of course hflve en NSAP. Often an end unit' will have only one NSAP, but for o vsriety of ressons, an end unit can have a plu-rclity of NSAPs. Distributlon units will normally also have NSAPs, so thot they ccn treated os end units for purposes ~uch as system control ~nd malnten-ance (system oana~ement).
When an end unit wents to send 3 packet to another end unit, lt constructs the pscket with a header or control sectlon includin~ its own NSAP and the NSAP of the end unit it ls sendin~ the p~cket to - thst is, the-source snd des-tinatlon NSAPs. In general, the trcnsport of ~ pocket throu~h the communico-tion network between two end units involves a number of distribution units.
Each unit (end units as well as distribution unlts) maintains some routln~
informotion, conventionally in the form of routin~ tables, which enables thst unit to-decide which of its neighbouring unlts it should forw~rd the packet to as the next hop in its path tG the ultimote destination. The routin~ t~bles ~ssoci~te each possible destination NSAP with some forw~rding information. Thelatter includes the identificstion of the link over which a packet containin~
particular destination NSAP should be forwarded.
If the link has multi-occess cspability ~such as 9 Locsl Area Network (LAN) link which permits many units to connect to the same link), then the forwarding informotion also inrludes the llAN identifier of the particular nei~hbourins unit on thet l~nk to which the packet should be forworded. (This LAN identifier is o layer 2 oddress, an~ is lndependent of the layer 3 ~ddresses, NSAPs, with which we ere primarily concerned.) In genersl, the routin~ tobles of the distribution units must maintain complete routin~ information obout ~11 units in the network. This is essentialfor the core network to be oble to pass (dot~) packets to all end units; if the core network does not have a record of the existence of an NSAP, it obviously csnnot locate it (sddress it) and so cannot p85S p9ckets to lt. The routin~
tables of the end units need not, howeYer. meintain complete lists of all the distribution units to which they ~re connected. It is desiroble for them to hold a re~sonable number of distribution unit addresses, but ~ sin~le distribution 20711~
unit address would be sufficient to 9110W the end unit access to the core net-work.
When an end unit wents to send a p~cket, lt looks in its routing tables for an entry with the destination NSAP. If lt finds uch an entry, it forwsrdQ thepocket usin~ the forwarding informotion ossociated with th~t entry ln the t~bles.
If it does not find a match, it chooses one of the distribution units present inits tables and forw~rds the packet to thet distribution unit. At the qsme tlmelt makes sn entry in lts routin$ tables for th~t NSAP with forwardln~ lnformo-tlon set to that of the chosen distribution unlt.
If the receivln~ distribution unit finds from its routin~ tables that the unit to which it should forw~rd the packet is on the same link ~s the packet was received on. then it may send a speciel redirect control mess~e to the source end unit, informin~ it that it c~n forward direct to the next hop unit for packets sent to th~t NSAP. The end unit receivin~ the redirect control messagethen upd~tes its routing tables entry for th~t NSAP with the forwardin~ infor-mation provided by the distribution unit.
A distribution unlt contalns o considerable amount of routing information in its routing tables. In principle,-an end unit need only contain the identi-fier of a sin~le distribution unit in lts routin~ tables. However, it is conven-lent for it to contain some routin~ information. This information will controlthe selection amon~ the distribution units in dependence on the destinstion NSAPs or, if the destin~tion unit is ~ neighbour of the end unit, it will contain the forwerdin~ informstlon necessory to send directly to the destinstion unlt wlthout pessin~ throu~h ~ distributlon unlt. Thus the routes to destinationc can be optlmized.
B~ck~round - rou~ng tuble m~intenance ond Hello mçss~ges There is a vsriety of ways in which the routin$ tables can be $enerated and maintained, The routin~ functions of end units are concerned only with the lniti~l hop of the path of a p~cket - from that end unit, to either the des-tinstion end unit or some distribution unit. The routing functions of distri-bution units c~n be divided into two cl~sses: those concerned with the finol hopln the pDth of s p~cket (from that distributlon unit to the destin~tlon end unit), 207~13~
~ 4 ond those concerned with intermedi~te hops (DU-DU hops, to further distribution unlts).
To deal first with DU~DU hops, there 9re m~ny ways in which routing infor-matlon for these hops m~y be m~int5ined ln distribution units. These units typically exch~nge control messages, formin~ a routin~ protocol, which contain routin~ information, and esch psrtlcipstes in a distributed routing al~orithmwhich ensures that eoch distribution unit will construct routin~ t~bles thst arecompatible with those in each other distribution unit. The compstibillty ensures that, for example, p~ckets will eventually be delivered to their destin~-tion ond will not repetitively circulQte around a closed loop of distribution units.
End units typically do not particip~te in the routin~ algorithm used by the distribution units, relying instead on their neighbouring distribution units to forw~rd packets appropriately. Thus end units typic~lly have much smoller and simpler routin~ tables thon distribution units.
We ~re primarlly concerned here with routin~ involving end unlts - EU-EU, EU-DU, and DU-EU hops. For this, the routin~ functions requlre knowled~e of the existence of neighbouring units. Such information may be confi~ured into e~ch unit by sn operator or lt may be learned dynamicslly through exchan~e of messa~es. The present system is concerned with networks in which this infor-mation is learned dynomicslly throu~h the exchange of control messages c~lled Hello messoges. Such messo~es are sent to neighbouring units but cre never forw~rded onwards through the network.
An end unit sends out EU-Hello messages to announce its presence to neighbourin3 distributlon units; ~nd similerly, a distribution unit s~nds out DV~
Hello messa~es to announce its presence to neighbourin~ end units. Option~lly,units of g psrticuler type cdn also note the presence of nei~hbourin~ units of the same type (e.g. end units can note the existence of other end units by listening to EU-Hello messsges~.
The Hello messa~es sre sent out st regulsr intervals, to confirm the con-tinuing presence of the sendin~ units. Ecch receiving unit contsins timers forthe various sending units which it is receivin~ Hello messoges from, to check that it continues to receive the ~ppropriate Hello messa~es; lf a timer tlmes out, 2071 13~
: - 5 -then the receiving unit deletes the correspDnding sendin~ unit from Its records.Esch unit is in ~ener91, of course, both a sendin~ unit and a receiving unit for Ha~ ~OSS~QS, The timer rste is norm~lly set to be lower thDn the rate ot which the associated type of sending unit sends Hello messc~es, so that on occ~sion31 lostHello message does not result ln the receiving unit deleting o sendin~ unit fromits records. If c distribution unit should dlsoppeor, that foct needs to become known to other units quickly, to ovoid packets being forwarded by mony unlts to nei~hbouring distribution unit thot hos stopped working; if on end unit should cease workin~, only p~ckets destined for that end unit will be lost. Distribu-tion units therefore normally send out Hello messsges much more frequently thsn end units.
Each EU-Hello mess~e includes the NSAPs of the end unit produclng it.
Any unit which receives an EU-Hello message records the NSAPs ln its routin~
tcbles. It also records, ogoinst each such entry in the tsbles, the forwordin$
information necesssry to send ~ packet to thst end unit. Norm~lly, the forw~r-din8 informotion includes ot least the identification of the link over which theEU-Hello message w~s received. If thot link ls o multi-occess link, then the l~yer 2 addressin~ info~rmotion necesssry to send a packet over that link to th~t end unit snd derived from the received packet m~y also be included in the for-wsrdin~ information, olthou~h it is possible thât this informstion could be derived subsequently by some other me~ns.
Each DU-Hello messoge will normslly include the NSAPs of the distribution unit producin~ it (though in some vorionts of this scheme, such informotion is not necess~ry and so is not included). Each end unit which receives these messoges records them in its routin~ t3bles, to~ether with the ssme forwording informâtion described obove.
In summ ry, then, the present type of communicstion system essentially involves two types of unit, the end units and the distribution units. The end units contsin routin~ tsbles which cont~in ~ rel~tively small ond possibly incomplete ~mount of lsyer 3 informotion re~ordin~ immediately cd~scent units;
the distribution units contoin routin~ t~bles which collectlvely contoin complete informotion about ~11 the units in the system ond routes between them.
6 2071~3~
The present invention The crux of the present invention i~ the provi~ion of a further type of unit which can be connected between end units and a distribution unit and which appears as a di~tribution unit to the end units and as an end unit to the diqtribution unit. We will call this new type of unit a hybrid unit (HU). A hybrid unit ha~
only one link to a di tribution unit, but can havs a plurzlity of links to which end units are connected (directly or via .further hybrid units).
The invention in its broad form i~ a3 follows:
In a communication network comprising end units (EU ) and distribution units (DUq) coupled together by communication links, each end unit having at lea~t one network service acces~ point (NSAP) or address which identi~ie~ the unit and each distribution unit including routing tables for routing data packets through the network between end units, the unit~ maintaining knowledge of their neighbouring unit~ by exchanging Hello messages which indicate the unit type and contain the NSAP (9) of tha end units, the DUY
collectively maintaining complete information about all NSAPq, a hybrid unit (HU) having one link (HUE link) on which it simulate~
an EU and a plurality of links (HUD links~ on which it ~imulate~
DUs, ~ending on the HUE link ~U-Hello message~ containing the NSAP~
of all EU~ which it receives on its HUD link~, and sending on the HUD links DU-Hello messages.
The links of a hybrid unit to which end units may be connected appear a~ links to a distribution unit to those end unit~; the link 6A 20711~0 by which the hybrid unit i~ connected to a distribution unit appears a~ the link to an end unit to that di~tribution unit. Thi3 combination of features means that the hybrid unit can be used in a communication ~ystem of the present type without requiring any modification of the existing units of the ~y~tem.
A~ described in d~tail hereinafter, a hybrid unit combines certain properties o~ end units and of diqtribution unit~. The hybrid unit provides some o the function~ of a distribution unit, but a major distinction b~tween the hybrid unit and a di~tribution unit is that the hybrid unit contain3 only a limited amount of routing information compared to a distribution unit.
The routing information of the hybrid unit i8 derived only from information obtained from its neighbourQ, and is similar in qcope to ~hat maintained by an end unit. In contrast, the routing information of a di~tribution unit contain~ all po sibls destinations in the network. The routing information processing which the hybrid unit ha~ to perform i~ thexefore much simpler than that which a distribution unit has to perform. The complexity o a hybrid unit is therefore closer to that of an end unit than that of a distribution unit.
The hybrid unit as 80 far de cribed may be of fixed configuration. It can however be made reconfigurable, either manually or automatically. Such a reconfigurable hybrid unit further enhances the flexibility of communication systems in which it is used, particularly when the syqtem changes (a3 the re~ult of either deliberate changes to its topology or failureq of part~ of the system).
6B 2 0 7 ~ 1 3 0 Since di~tribution unit~ are complex, it i~ desirable to minimize their number. If thers i8 a modest number of end unitR in an area which is rea~on-20711 3~
.
ably compect ~eogrsphicslly, thls c~n be achieved by coupling the end unitsdirectly to a single distribution unit. If the number of Pnd units in such an sree is l~rse, this technique becomes inconvenient. is the number of ~orts available on a distribution unit is generally fairly smsll. However, a 13r~se number of end units can be connected to a sin~le distribution unit through a multi-access link such as a LAN, as descrlb~ obove.
If the end units are widely dlstributed $eogrophically, however, then it is $enerally not feasible to connect them to a single distribution unit elther directly or by a LAN. With o conventlonol system, they must be grouped into ~eographicall~J small groups, each of which has a sepsrote distributlon unit.
The hybrid unit effectively provides an altern~tive w~y of connectin~; end units in such a situatior.. Esch geo~raphically smsll group of end unlts is connected to a hybrid unit, and the hybrid units are in turn connected to o sin~le distribution unit.
BPIE~ D~ IO~I OF D~L7~
Two types of hybrid unit, a simple (non-outoreconfigurable) unit and an autoreconfigurable unit, embodying the invention will now be described, by way of example, with reference to the drawin~;s, in which:
Fi8. 1 is a block dia8ram of a communication system includin~ hybrid units;
Fi8. 2 is a block dia~ram of an end unit, wlth particular reference to its routing tsbles;
. .
Fi$. 3 i4 a block dia8ram of ~ distrlbution unit, with particulcr reference to its routing tables;
Fig. 4 ls e block diegram of a simple hybrid unit, with psrticular reference to its routing tablesi Fig. 5 is 9 block dis6rsm of the components of an Dutoreconfi~urable hybrid unitadditional to those shown in Fi~. 4;
Fig. ô is a $eneral flow dia~ram of the operotion of the hybrid unit of Fi~;. 5;and ` 2071130 Fi~s. 6A to 6C show the blocks of Fi~. 6 in more detail.
General system Fi~. 1 shows a communication system includin~S four distribution units DU1 to DU4. Distrlbution unit DU1 is connected to the LAN link LANI, which ~lso has three end units EU1 to EU3 connected to it. Distribution unit DU2 is con-nected to a further end unit EU4 ond olso, through seporote links, to the LAN
link LAN1 and o further LAN llnk LAN2. Distribution unit DU3 is connected to three further end units EU4 to EU7; the end unlt EU4 i5 also connected to dls-tribution unit DU2 (as just noted).
The system as described so fsr is conventional, ond shows o variety of features of convention~l systsms (slthou~h many systems do not include all th sevarious feotures).
End unit , Fi8. 2 is a block dio~rom of the routing tables and ossociated components of an end unit (say EU1). The main components are a local NSAP list store 10;
a control messa~e receiver unit 16, whose purpose is receiving control messa~es;two routin~; tables, e distribution unlt table 19 and an NSAP table 20; end a route look-up en~ine 22.
:
The local NSAP list store 10 contoins the NSAP(s) of the unit, snd is coupled to an EU-Hello messa$e ~enerotor 11, which is olso fed from on EU-Hello timer 12. At sultoble intervals determined by the timer 12, the messa~e gener-ator 11 assembles EU-Hello messages, containin~ the NSAP(s) of the end unit, andtransmits them on the links from the end unit (e.~. a LAN, or links such as thatfrom end unit EU5 to distribution unit DU3 or those from end unit EU4 to dis-tribution units DU2 and DU3).
Eoch location in ths distribution unit table 19 consists of two sections, a timer section 19-T and a distributior, unit forwardin~ information section 19-E.Any incomin~ DU-Hello message from a dlstribution unit neighbourin~ the end unit is received and proressed by the receiver unit 16. This unit writes, in section l~F of table 19, in~ormation obout the exiQtence of the distribution unlt, together with the ~orwarding informotlon neeess~ry to send o pocket to the 207~130 g dlstribution unit. As a result, the tabie 1 g will gredu~lly accumulate the identifiers of the distribution units coupled to the end unit. The number of entries in the tabla may be restrlcted so that only the most recentlv hesrd'fromdistribution units are contcined in the table, for example.
The unit 16 also writes into the timer section I 9-T of each locction In the distribution unit table 19 the time tfrom a deletion clock unik 17) when the last Hello message for that location was recelved. The deletlon clock unlt 17 constantly comp~res the stored times ln the tlmer section 19-T with the current time, and erases the entry from the dlstribution unit table lf the time difference exceeds a preset value. (This value moy be derived from g field in the DU-Hello, so that it can be different for different distrlbution units in the table.) Thus the contents of the distrlbution unlt table are maintained to match the active distribution units currently neighbourin~S the end unit.
I~ch location in the NSAP table 20 consists of three sectlons, a timer section 20-T, an NSAP section 20-ID, snd s forwarding informstion section 20-F.
For each NSAP contained in the table, section 20-F contains forw~rdin~ informa-tion for use when sendin~ packets destined to that NSAP. A deletlon clock unit 23 operates, in conjunction with the timer section 20-T of each location inthe NSAP table and the wrltin~ lnto those sectlons of the last update times, to delete obsolete entries from the tsble.
When the end unit wants to send a data packet, it f irst uses the route look-up en~ine 22 to look for an entry in the NSAP t~ble 20 whose NSAP (in sectlon-: 20-ID) matches the destin~tion NSAP for the pecket. If such an entry exists in the NSAP table, the end unit uses the associsted forwording inform~tion in section 20-F to send the packet to the next hop neighbourinE~ unit.
If there is no entry in the NSAP table for the destination NSAP, the route look-up en~ine selects an entry, e.~. at random, in the distribution unit table 19.
It adds a new entry to the NSAP table 20 with the NSAP section 20-ID set to the destination NSAP of the packet, the forwardin~ sectlon 20-F set to the for-wardin~ information extracted from the selected information in the di,tribution unit table 19, and the timer section 20-T set to a suitable value.
If a distribution unit respond~ by sendin~ a dlstribution unit redirect messo~e to the ori~;inatin~; end unit, this ls passed by the control messa~se 2 ~ 3 ~
~o -receiver unit 16 to a distribution unit reclirect en~ine 21, which changes the forwarding inform3tion in section 20-F of the NSAP table 20 to that specified inthe dietribl tion unit redirect message and resets the ~ssociated timer in section 20-T (to 9 preset v31ue). When the end unit st~rts up, the routing tsble 20 is initially empty. As packets are sent out to different destinations, so entries graduslly accumulate in this table.
E~ch time an entry is mode in the NSAP table 20, the timer section 20-T
for that entry is updsted. Entries in this t~ble eventu311y time out and are deleted by the deletion clock 23. They are then reformed ss described above when new packets are sent by the end unit.
The size of the NSAP table 20 is usually limited. If elther the route look-up en~ine or ths distribution unit redirect engine needs tn make a new entry in this tsble and it ls full, then an entry is deleted; The entry for deletion may be chosen at rondom or based on the velue of its ~ssocieted timer section 20-T.
As noted sbove, it is possible for end units to listen to EU-Hellos. This provides another source of information for msintaining the NSAP table 20.
This will, however, involve fairly substantial processing overheads, and is liable to fill up the table with NSAPs which the end unit is unlikely to want to send psckets to, so in practice this is not usually done.
Distribution unit Fi8. 3 i9 0 block diagr~m of the rout1ng tables end associoted components of a distribution unit (say DU1). The distribution unit has a local N~AP table 31, a remote NSAP table 37, a control message receiver 32 (which is mainly for EU-Hello messages), ond a control messa~e generator 33 (which is for DU-Hello messages and redirect messsges).
The DU-Hello messages are genersted in unit 33. These are transmitted on all links of the distribution unit at suitable intervals, as determined by a timer 34.
The local NSAP table 31 contains ~n NSAP section 31-ID, a timer section 31-T, ond o forwardin~s inform~tion section 31-F. Thls table contain~ the NSAPs _~ 1 ~ :~ Q ~
_~
_ _ _ ~ I
of er"i urlits wr:~-h are nei~hbours ot` the distribution unit~ ElJ-Hello messâges 3re rP eive.i b; :.r,it - ar,~ P ~ISAP¢s) arl-l as oci3ted forwarding ir,formatior, ~' are written into the appropriate ecti:)ns :f the tJ~le ~1l alon~ with the tlnle :;
receipt. Thus he table ~11 will gradu311y accumulate thê identifiers c.f the encl ~F units :oupled t.-. ~ ;-,e distribution unit. A d-letion clock unit ~c; rr,aintains r these entrie .urrent in the samP W2 ~ as the deletion clock unit 1, c f the end unit shown in FiF. '.
The distribution unit also contâins a remote NSAP table 37 which like table ~1 contain entries associatin~s NSAPs (in ,ection ~7-ID) with forwarding inform3tion (in ,e.tion 3,-F~. Entries in this table are made and deletêd by a routing control -ngine ~ which implements the distributed routing algorithm used by all the distribution units. This table contains the NSAPs of end units which are not neighbours of the distribution unit. ~In practice. the orgâniza-tion of this table may e;~ploit the generâlly hierarchical nature of NSAPs.!
When the distribution unit receives a packet it matches the destinetion NSAP of the packet against the entries in sections 31-ID and 37-ID of the routin~s tables using the route look-up en8ine 3fi. The forwarding informatior.
is obtâined from the appropriate routing table. and the packet i5 forwarded towards the destination. If the de-tination ~ISAP is in the local table 31. the destination end unit is 3 neighbour of the distribution unit. If it is in the remote table 3, then the end unit is not a nei~shbour of the distribution unit so the forwarding information will cont3in details of the next distribution unitto whi h the pack-t is to be forwardeà.
If the forwarding information used by the distribution unit results in the pa-ket being forwarded to another unit connected to the same link as that on whic.' tr.e p~ l_r WâS r cei~ d r,- -u;.irg 'ol~-a? -n~ir,e ;f~ . ner3t-s a reiire:t rr)ess2g -oritdir.irl;r th- destir.3ti~rl NSAF 3n] ~he f~.~ wardir,~- inform3-: n n,atcr,ir,g that `;SAF in the routin. ~Dle_. ar,d sends the rr,es iage t; tne 5~ ir~e end unit.
The operation of the systern as described 50 far is onvêntional 3nd shows a variety of features of conventional systems ~although many systems do not include all theese various features).
2~711 30 Hybrid unitq - ~eneral Returnin~; t,~ Fi~. 1. the syst~m a1so includ~s ~ distribution unit r~U4, wi+h a nei~hbourin~ end unit EU7 snd ~lso two nei~hbouring hybrid units HU1 snd HU2. Hybrid unit HU1 has fl nei6hbourinE~ end unit EU8 ~nd ls also attached t LAN link LAN2. Hybrid unit HU2 has a neighbouring end unit EU11 and a further nei~hbouring hybrid unit HU3. Hybrid unit HU3 has two nei$hbouring end units EUl2 and EU13. These parts of Fi~. 1 show a variety of ways in which hybrid units may be used (~lthou~h mony systems will not include all these vorious features).
The hybrid units are strictly hierorchlcal, with a sin~le superlor link snd a plurality of subordinates allowed. On its superior link, ~ hybrid unit appeq~C 95 an end uni+; thus hybrid units Hlll and HU2 sppeor 9S end units to distribution unit DU4, sending EU-Hello messages up to it and receivin~ DU-Hellomesss~es from it. To its subordinates, o hybrid unit appeors ~s a distribution unit; thus hybrid units HU1 to HU3 oppeor as distribution urits to end unlts EU8 to EU13, sendin~ out DU-Hello messa~es to those end units and recelving EU-Hello mess~ges from them.
In principle, HU1 could have more than one port appearin~ as an end unit (rather like end unit EU4>. This would result in multiple paths from the end units subordinate the hybrid unit. In the csse of a hybrid unit, this would produce no significant sdvanta8e and would result in various complications, so we prefer not to implement this possibility.
, .
Hybrid units can be coscAded, as shown by hybrid units HU2 snd HU3. The same hierarchic rules apply, so that hybrid unit HU2 appears as ~ distribution unit to hybrid unit HU3 and hybrid unit HU3 appeers as ~n end unit to hybrid unit HU2, with EU-Hello messages passing upwords from HU3 to HU2 ~nd DU-Hello messa~es p~ssing from HU2 to HU3.
A hybrid unit performs both Hello messa~e and dats packet routing func-tions. Considerin~ first its response to data packets, it forwards these in bothdirections. It performs a routing function on these, forwardin~ packets coming from the distribution un~t above it on to the ~pproprlote link to the destination unit, and forwording packets comin~ from the end units below it either on up to the distribution unit above it or directly out downwords to the destinatlon unit.
2~`13~
Considerin~ its routln~ control functions, the hybrid unit collects the NSAPs of its subordinate end units received throu~h EU-Hello mess~ges ~inclu-din~ thocr whirh are- indirectlv subordinate through further hvbrid units). anclappears as a sin~le ordinsry end unlt ~albelt with sn unusually large number of NSAPs) to its superior unit (whether a further hybrld unit or a distrlbution unit) by psssing on, in EU-Hello messages whlch lt sends to it3 superior, all the NSAPs it hes collected in this way.
Simple hybrid unlt FiS. 4 is a block dlegram of the routing table and assoclated components of a simple (i.e. non-3utoreconfl$urable) hybrid unit (say HU1). This contalns three routing tables: a distribution unit table 45 and an NSAP table 54, corres-pondinP resp.oçtivel~f to the distribution unit table 19 and NSAP table 20 of the end unit of Fi~. 2, snd an end unlt table 49, correspondin~ to the local routingtable 21 of the distribution unit of Flg. 3.
Considering first the upward flow of messages, an EU-Hello message receiver 53 receives EU-Hello messa~Ses from end units and posslble hybrid units lower inthe hierarchy, and feeds these messa~es lnto the end unlt table 49. This table hes three sections, section 49-ID for the NSAPs contained in the incoming mes-sages, section 49-F for nssociated forwarding information, and section 49-T for the times of receipt. A deletion clock and timer unit 52 provides these times and deletes entries in the table which are not updated by fresh EU-Hello mes-sages at suitable intervalQ. The teble 49 must be lar~e enough to retain all NSAPs-of 911 the end units below the hybrid unlt. (This is like the routing tables of the distribution units, and in contrast to the routin~ tables of the end units.) An EU-Hello generator 42 i~ coupled to the table 49 and to 9 local NSAP
list store 43 which stores the NSAP(s) of the hybrid unit itself. At intervals determined by a timer 41, this generates EU-Hello messa~es which include all theNSAPs in end unit table 49 and those in store 43 and transmits these to the distribution unit or superior hybrid unit.
Considering now the upward flow of data paclcets, if the hybrid unit receives a data packet coming up to it from an end unlt or subordlnate hybrid unlt, the route look-up engine 50 inspects the header of the packet to determine ~7~39 whether there is 9 match between its destinotion NSAP and any N5AP in the end unit tsble 49. If there is, then the hybrid unit forwRrds the packet on the appropri3ta output link using the forw~rdins information assocista:l with th~t NSAP in end unit table 49.
If there is no match in the end unit table 49, the route look-up en~ine 50 looks for a match between the destination NSAP of the p~cket and any NSAP in the NSAP table 54. If there is, then the hybrid unit forwards the packet on the sppropriate output link using the forwarding information associated with that NSAP in the NSAP table 54. T~ble 54 ls updated by redirect messa~es received by a redirect messa~Se receiver 51 from a distribution unlt or superiorhybrid unlt and by a deletion clock unit 55.
Tf there is no mstch in either of these tables, then the hybrid unit for-wards the packet up to a distrlbution unit or superior hybrid unit using the some algorithm as that used by the end unit. The route look-up engine 50 enters the destinstion NSAP into the NSAP tsble 54 to~ether with the forwardin~Sinformation associated with the distrlbution unlt.
The route look-up englne 50 may ~lso ~enerate updating redirect control messages like those 8enerated by the distribution units, as discussed above, sendlng these to the end units and subordinate hybrid units.
Considering now the downward flow of data packets, the hybrid unit inspects the header of a pscket received from a distribution unit or superior hybrid unit, matches~.. the NSAP against the NSAPs ln the end unit table 49, and forwards thepacket downwards according to the forwardin~ inform0tion associated with th NSAP in table 49.
Considering finally the downward flow of control mess~es, the hybrid unit receives DU-Hello messages sent by superior distrib~Jtion units or hybrid units and these ere processed by the DU-Hello message receiver 44 and stored in the distribution unit table 45 (in exactly the same way as an end unit processes these messages). On those links to subordinate units (hybrid unit;, or end units), the hybrid unit periodically sends DU-Hello messa~es using the DU-Hello gener~tor 47 controlled by the DU-Hello timer 48. .
2~7~
The hybrid unlt also receives distribution unlt redirect messages through the distribution unit redirect messa~e receiver 51 and modifies the information in the N!~AP table F~4 based on the NSAP 3nd fc rwardin~ information received in the distribution unit redirect messa~e.
The hybrid unit thus appear as on ordinary distribution unit to the end units ond ~ny subordinate hybrid units, and os an ordinory end unit to the superior distributlon units or superior hybrld units.
It will be noted that if there should be a foilure of the superior unit or units of a hybrid unit (i.e. of the distribution or superior hybrid unit or units), the port of the network includinES ond below the hybrid unit can continue to operate. The hybrid unit con continue to forword pockets between the end units below it.
In ~eneral, each hybrid unit has severol links. One of these links is that connected upwords in the hierorchy, to o distribution or a superior hybrid unit, and behavin~ as on end unit link, and we can term this an HUE link. The remoinlng links oll behove os links of e distribution unit, ond we con term themHUD llnks.
Autoreconfi~5uration We have assumed so far that the HUE or HUD noture of the various links of the hybrid units is predetermined. In proctice, it is convenient for the nature of the links not to be predetermined. This allows the hybrid unit to have its links connected arbitrorily. If this is the case, then the hybrid unit must be confi~ured so thot the appropriate link becomes its HUE link and the remainin~ links become HUD links.
The simplest form of configuroble hybrid unit is one in which the configu-ration is operotor determined. This may require dlrect manual settin~ of the links, or it may be performed by control messa~es ~eneroted by the operator but sent throu~h the network to the hybrid unit.
In large ond complicoted systems, there may be more thon one possible woy to configure some of the hybrid units, or it m&y be desirable or necessary to change their confi~urations in the event of a network change (which may be a 2~7~ 3~
~ 16 --permanent addition to the network or deletion of some part of the network, or may be a temporflry change resultin8 from a temporary failure or recovery from a failure, fc~r example of a link~. s If, in the Fi~S. 1 system, distribution unit DU4 were to fail, end unit EU8 would become in~ccessible. It would therefore be desirable to reconfi$ure - hybrid unit HUl to presert an HUE link to the LAN link LAN2, and hence to distribution unit DU2, so m~king end unlt EU8 occessible agoin. Such reconfi3-urotion can be achleved by operotor control, ~ust ~s with initial confi~uration.
It will be reclized that the fcilure of distributlon unlt DU4 makes end units EU7 and EUl 1 to EUl 3 inaccessible to the rest of the system, ond this connot be rectified by ~ny reconfi~uratlon. End units EUl 1 to EU13 remoln in c4mmunication with each other, however, Vi3 the hybrld units HU2 and HU3.
The Fig. 1 system could be configured with hybrid unit HUl presenting an HU llnk to distribution unit DU2 via LAN link LAN2, and bein~; effectively dis-connected from distribution unit DU4 even if that unit were fully functional.
This could result, for exsmple, if dlstribution unlt DU4 recovered after its failure.
Direct manual reconfi~uration is often f~r from convenlent, because the part of the system where 3 failure occurs may be remote from an operator cap-able of performin~ the reconfi~Suration. Reconfiguration by remote operetor control is also often not satisfactory. It may be difficult or even impossible for a remc,te operator to determine the precise nature of the system failure; ~nd bec3use the system hss under~one o feilure, lt msy be difficult or impossible ~chleve the optimum reconfiguration remotely.
The hybrid units therefore preferably include autoconfiguration meflns, whereby they confi~ure the HUD or HUE nature of each of their links automatl-cally.
If a hybrid unit has no hybrid unit as neighbour, autoconfiguration requires only that it should find a distribution unit to which it can present ~nHUE link. Thus hybrid unit HUl con autom~tlc~lly rsconfi~5ure itself, in the event of ~ failure of distrlbution unit DU4 or its link to that unit, simply by 2071~
~ 17 --makin~ lts link to LAN link LAN2 an HUE link, so becoming subordinate to dis-tribution unit DU3.
If a hybrid unit has a hybrid unit neighbour, however, then it msy be desircble to reverse the HUD-HUE direction of the link between the two hybr$d units; and if both the hybrid units have distribution units a5 neighbours, then this mey be necessery. Further, the possibility of a circulating peth for data packets must be ~voided. (This could hsppen if, for exemple, there are three hybrld units ell connected together.) In generel, therefore, en autoreconfi~uro-tion algorithm is required which results in the system remaining connected as far as possible while cvoidin~ inconsistency.
There are o; course various possible 13eneral algorithms for autoconfiguring a network containin~ hybrid units. The ma~or ob~ects to be achieved in genercl can be defined in various ways, such as:
there should be no closed loop of hybrid units with HUD to HUE connec-tions;
2 no hybrid unit should have more than one HUE link; and 3 a~ many hybrid units as possible should have HUE links.
If connection to the network (i.e. to a distribution unit) is possible, then condition 3 means that every hybrid unit has en HUE link, and condition 1 then means that all hierarchically upwerd paths through hybrid units must eventually terminete at a distribution unit. If connection to the switching network is not possible, then these conditlons result in precisely one hybrid unit not having an HUE link, i.e. becoming the master hybrid unit, wlth all the other hybrid unit forming hiererchically upward paths to that mester hybrid unit.
We will describe a mechsnism using c spanning tree algorithm which permits erbitrary topolo6ies of distribution units, hybrid units, and end units. Simpli-fications of this mechanism are possible if the topology of the subnetwork of neighbouring hybrid units, plus the distribution units end end units neighbour-ing the subnetwork, is restricted such that there are no alternative paths between any two units, and there is only one link from ~ connected set of hybrid units to a distribution unit.
2~7~130 ~ 18 --To permlt autoconfi~uration, a new control message is defined, termed e HU-Hello. This allows neighbourin8 hybrid units to reco~nize each other as hybrid units, snd rc~ntains some inform3tion wh~ch aLlows hybrid units tc) USa 3 spansin~
tree algorithm. Hybrid units use the algorithm by combining information from the HU-Hellos they receive with their own local information.
The spanning tree algorithm ~enerates a singly-connected logical topology for the subnetwork from the srbitrQry physlcal topology, ond in sddltion forces each hybrid unit to a8ree with its hybrid unit neighbours the HUD/HUE statss of its links so that the hybrid units of the subnetwork act together correctly.
The algorithm selects e single hybrid unit as the root of a tree spanning the subnetwork, choosing for this purpose the hybrid unit with highest priority (as described below).
The priority scheme is desi~Sned so that the alE~orithm will operate in all hybrid unit ~ubnetworks, whether or not they hove distribution unlt nei~hbours.
If o hybrid unit has a distribution unit- neighbour, its inherent priority is replaced by a boosted priority which is hi8her than the inherent priority of anyhybrid unit. In the presence of distribution unit nei~Shbours, therefore, the e40rithm always selects as the root hybrid unit a (single) hybrid unit which hasa distribution unit nei~hbour.
The spanning tree Qlgorithm operates over the whole of the subnetwork of hybrid units; processin~ is corried out in the individual hybrid units, with communication between them for the purposes of the al~orithm bein8 by means of the HU-Hello messages. The operation of the ~lgorithm is thus distributed amon~ the various hybrid units of the subnetwork. The processin~ in each hybrid unit conslsts of a sequence (cycle~ of operations which is repeated each time the state of any nei~hbour of the unit chan~ses in a significant wey. Once a cycle has been initiated in a unit, that cycle will automatically proceed to completion. The eventual result is that the hybrid unit with the highest priority is selected as the root, to which all other hybrid units become inferior.
with a simply connected connection pattern. This may involve makin8 certein links effectively inoperative for the transfer of data messa~es.
The operation of the al~Sorithm i5 initiated whenever any change in the st~te of the system - e.~ a link or ~ distribution unit or a hybrid unit becoming operat1ve or inoper~tive - occurs. This may be detected by e hybrid 2~71 13~
,9 unit noticin~ some change - e.~. a link becoming operative or becoming inoper~-tive.
Esch cycle of the elgorithm in a unit consists of three successive sta~es, I
to III, shown as blocks 70 to 72 ln Fi$. 6. In st~e I, a priority, a root cost, and e designation are determined for each link. In stoge II, the hlghest priority link is determined~ In stage III, the state of each link is set to HUD
or HUE from the informetlon calculated in sts~es I ard II. The three sts~eC
ore performed in successlon, and the unit then enters a woitin~ stote, block 73,ond remains in thet state until o chsnge occurs. The chonge may -be ony of thestote variobles used os input to the spannin~ tree olgorithm such ns the opera-tional state of a link or the contents of o neighbour's HU-Hello messoge.
Hybrid uni~s send HU-Hello messs~es on all their links~ Nei~hbourin~ d1s-tribution units and end units l~nore these mess~ges - only neighbouring hybrid units receive and oct upon their contents. The contents of on HU-Hello mes-so~e ore link-dependent, Qnd include 9 priority field and a cost field. The priority is set to the highest priorlty from all link priorities ~excluding the link over which the HU-Hello message is to be sent~ ond the priority of the hybrid unit itself. The cost field contains the accumulated cost of the path from a hybrid unit to the root hybrid unit; this is the sum of the costs of eochof the links on the path from the hybrid unit to the root hybrid unit.
The spanning tree algorithm requires various choices to be mode between different çntities of the same kind (such as hybrid units, or links). For these choicesm eoch of the entities hoQ a value termed 9 priorlty, and the choice is m~de by selecting the entlty wlth the hi~hest priority. For convenlence, theseprlorities ore defined as unsi~ned inte~er numbers with lower integer values having higher priority. Also for convenience, seporote priorities ~re combined together in various ways to perform priority comporisons.
In perticulor, the priority field of an HU-Hello message is a concatenation of three subfields which are, in descendin~ order of si~nificance: DUbit, ~ sin~le bit which is O if the hybrid unit has a nei~hbouring distribution unit ~nd 1 otherwise; e hybrid unit priority value which is configureble by the operator;
and a unique identity value HU-ID. The highest prlority is token os the lowestvdue (treating the combined bit field~ as unsigned number4~. We will use the 2~1130 not~tion <x,y,z> for ~ prlority formed by concotenating the fields x, y, and z, with x bein~ the most significant field~
The hybrid unit priority value can be used to manually configure 9 desired hybrid unit to become the rsot in the absence of ~ny distribution unlts attachedto the subnetwork; the HU-ID field must be unique for eoch hybrid unit in the subnetwork.
Autoreconfigureble hybrld unit . .
Fi~S. 5 is a block dia8ram of the circuitry (additionel to that shown in Fi~.
4) of a hybrid unit with outoconfi~uratlon. The main componentq are: 3 set of re~isters 61; a link teble 62; e nei~hbour table 63; ~utoconfiguration control lo~si^ 64; and an HU-Hello ~enerator 66 which is controlled by a timer 6~ rnd periodically sends HU-Hello messages on each of the links of the hybrid unit.
Deletion clocks 68 and 69 operete in con~unction with timer sections 62-T and 63-T in tables 62 and 63 to remove entries from these t~bles if no messages ~re received to update them within suitable time limits. The operation of the spsn-ning tree ~l~sorithm applied ot each hybrid unit csn be described ~long with this circuitry.
HU-Hello messages received by an HU-Hello receiver 67 are used to maintain information about the neighbouring hybrid units ln the hybrid unit neighbour table 63. The inform~tion consists of: the identification of the link on whichthe nelghbour exists (link ID). held in e section ô3-LID; the priority value contalnéd in the HU-Hell~ messa~e, held in e section 63-PR; the cost value con-tained in the HU-Hello message (the reported cost of the p3th from the neigh-bour hybrid unit to the root hybrid unit), held in a section 63-CST; the identi-fier of the sendin~; hybrid unit contained in the HU-Hello mess~ge, held in a section 63-ID. ~On a multi-access llnk such as a LAN, there may of course be many neighbouring hybrid units).
The hybrid unit elso malntains a link t~ble 62. Each entry in this table records information 3bout one of the links of the hybrid unit. The link ID, in section 62-LID, is the link identifier (unique to the link within the hybrid unit~.
The cost, held in section 62-CST, is o v~lue set by the operrtor to determine a preference for choosin~s one link over ~nother in the spannin~ tree dgorithm.
The distribution unit nelghbour, held in section 62-DU, is o 13Ooleon value (s - 21 ~
sln~le blt) set by the DU-Hello receive lo$ic 44 of Fi8- 4 to I ~true~ if the distribution unit nei$hbour table 45 of Fi~. 4 contains an entry for a distrlbu-tion unit on that link. and to n (false! otherwise.
Each link entry in the link table 62 includes a link priority section, ô2-LPR. The m~nner in which the link priority i5 determine~ is described below.
Esch link entry also contains a Hello-messa~e priority, held in section 62-HMPR,which Is the hi~shest of the link priorities of all other links in the link tebla and the prlority of the hybrid unit itself. The HU-Hello generotor 66 periodi-cally sends, on each of the linkQ of the hybrid unit. HU-Hello me sagec, the contents of which are link-dependent; in particular, the priority field of ~n HU-Hello messe$e is set to the priority value 62-HMPR from the appropriste entry in the link table 62. and the cost field is set to the velue in the re~ister 61-RC~T.
The set of registers 61 include~ a hybrid unit identifier re~ister 61-KUID
and a hybrid unit prlority reE~ister ôl-HUPR. Register 61-HUID stores the identifier HU-ID of the hybrid unit. This must be unique emon~ all hybrid unlts in e subnetwork; it may be congi~sured by the operator or -be derived fromsome other unique property of the hybrid unit. Re~ister 6 l-HUPR stores the hybrid unit priority of the hybrid unit. This is set by the operator to deter-mine a preference for one hybrid unit in the subnetwork over others in choosinE~the spanning tree root.
The other three re~isters in the set of registers 61 are set by the control lo~ic 64, which elso sets values in the link table 62. The control logic 64 implements the spenning tree algorithm. which consists of three successive sta~ses. The operetion of the control logic 64 c~n be tri8~ered in response to eny chan~e in any o~ the values used AS inputs to the lo$ic from any of the table~.
~ efore the first executlon of stage I. the system is initialized by settin~the re$ister ôl-RCST to O and, in the link table, section 62-HMPR to the value <1, 61-HUPR, ôl-HUID> for each link.
.
2B7~30 St~e I
.
In sta~e I, the control logic t;4 determinec, for earh link, a priori~y, 3 root cost, snd a desi~nstlon, and sets these in the entry for that link in the link table 62 (in. the priority section 62-LPR, the root cost section 62-RCST, ~nd the design~tlon section 62-DES respectively). The Deslgn~tion field 62-DFS for each link is used on multi-cccess links to select one hybrid unit only to act ~sDU on th~t link in the case where more thon one hybrid unit m~ly compute the link to be ~wey from the sp~nnin~ tree root. The hybrid unit with highest priority on the link will set Desi~nation to l (true) and ~rt e i o DU, while all other hybrid units on the link, for which the link is flW~y from the spsnning tree root, will set Desi~n~tion to 0 (felse) and send ond receive only HU-Helloson th~t link.
For eech link in the t~ble 62, the control lo~ic finds the hi~hest priority from the set of priorities derived from the priorities of ~11 nei~hbours on the link ~obtsined from the Nei~hbo~lr teble 63), ~nd includinE~ the priority sent ln HU-Hellos on this link by the hybrid unit itself.
Fig. ôA is ~n informsl flow di~gram for stsge I. The first block, block 80, ls fl decision block which c~uses the links listed in the link table 62 to be processed in sequence. For each link, the appropri~te sequence from the blocks 81 to 86 is performed; control then returns to block 80, and the next link is processed. When ~11 the links h~ve been processed, sta~e I is complete, and the N exit from block 80 ls taken, to st~ge II (Fi~. ôB).
.
After a link h~s been selected in block 80, the first operations performed are those of block 8l, initifllizing the link volues for that link. Thus the link design~tion 62-DES is set to l (true), the link priority 62-LPR is set to the priority for the Hello mess~es 62-HMPR. end the link root cost 62-RCST is set to the root eost 61-RCST in register set 61.
The distribution unit number 62-DU for the llnk is ther, examined in block 82. If this is l, indicatin~ that the link hfls 9 distribution unit neighbour,then block 83 is performed. In this block, the link priority 62-PR is set to <0, 61-HUPR, ôl-HUID>. The root cost for the link, 62-RCST,is set to the cost for the link, 62-C3T. Fin~lly, the designatlon for the link 62 DES is set to 0.
2~7~1~0 If the distribution unit num~er 62-DU is found to be 0 in block 80, then block 84 follows. This deçision block causes the hybrid units listed in the nel~hbour tsble 6~ for th~ current link to b~ process~d in sequence: wheri 911 hsve been processed, the operation returns to block 80 for the next link.
Block 84 is followed by block 85, which tests whether the priority fleld <62-PR, 62-RCST, 61-HUID> ls ~re~ter than the priority field <63-PR, (63-CST +
62-CST), 63-ID>. If it is not, then no ~ctlon is tsken, snd the system returnsto block 84 ~nd ~dvences to the next hybrid unlt for the link. If it is, then in block 86, the link priority 62-PR i9 replaced by the hybrid unit priority 63-PR, the llnk root cost 62-RCST is set to the sum of the link co~t 62-CST of thls link and the root cost 63-CST reported in HU-Hellos from this neighbour, ~nd the link desi~nation 62-DES is set to 0; the system then returns to block 84snd advsnces to the next distribution unit for the link.
Thus the links are processed in sequence, and for each link with hybrid units ~s nei~hbours, the processin~ is for those hybrid unlts in sequence. As result of sts$e I, the control logic 64 sets the value of priority 62-PR, root cost 62-RCST, snd Desi~n~tion 62-DES for eoch link in the link t~ble. When this process is complete. st~e I is finished.
Sta~5e II
After the priorities, root costs, and desi~nctions of 911 the links hsve been determined in sta~e I, stoge II follnws, which c8~in pro~esses the links in sequence. In this st~e, the control logic 64 determines the highest priority link - i.e. the link with the hl~hest priority 62-PR ~nd root cost 62-RCSI'.
If more thcn one link hes the s~me highest priority, the link with the highest link ID (section ô2-LID of the link entry in t~ble 62) is chosen. The link ID
of the hiEShest priority circuit is set in the root link ID re$ister 61-RLID. Ifthe hybrid unit itself hss 3 hi~her priority th~n the priority of sny of its links, then the root link ID re~ister is set to 0. This me~ns th~t there is no root link, and the hybrid unit is the spsnnin~ tree root in ~n isol~ted subnet-work. The hl~shest priority is set in the cslculated priority ree;ister 6 l-CPR.The root cost for the hi~hest priority link (or zero if the HU is the spannin~
tree root snd h~s no DU nei~hbour) is set ln the re~ister root cost 61-RCST.
2-~:L1 3~
~ 24 --Sts~e II (Fi~. 6~3) storts with an initi91izing block 90 whlch initializes both the hybrid unit root link ID in re~ister 61-RLID snd the hybrid unit root cost in reo~ister 61-RCST to 0 This is followed by block 90. ~ decision block which c~uses the links listed in the link tcble 62 to be processed in sequence. For each link, the sppropri~te sequence from the blocks 92 to 98 is performed: con-trol then returns to block 90, ~nd the next link ic processed. When all the links h~ve been prccessed, st~ge II is complete, and the N exit from block 90 istsken, to stcge III (Fi$. 6C~.
Block 92 tests whether the link priority 62-PR is equ~l to the Hello-mess&ge priority 62-HMPR for HU-Hello mess~e~ sent on thst llnk. If it is, nothinES further is done, end the system returns to block 91 to step on to the next link. If it is not, the next block is 93.
Block 93 compares the priority field <61-CPR, 61-RC5T> of the hybrid unlt with the priority field <62-PR, 62-RCST> of the link. If the result ls U~ the hybrid unit's prlority field is less th~n the link's priorlty field,~ i then no action is t~ken, and the system returns to block 91 to step to the next link. If the two fields ~re equol, then checks Qre made to see whether the hybrid unit's link identifier 61-RLID ls O ~block 95) and whether the link's identifier 62-LID is less thon the hybrid unit's root link identifier 61-RLID
~block 96). If the hybrid unit's link identifier is O or is gre~ter than the link's identifier, then then no action is taken, ~nd the system returns to block91 to step to the next link. If the hybrid unit's link identifier is ~reDter th~n 0 aDd less than the link's identifier, then it it set to eqtlal the link's identifier (block 97), and tho system then returns to block 91 to step to the next link. Fln~lly, if the hybrid unit's priorlty fleld is ~Sre~ter th3n the link's priority field, then the hybrid unlt's cost priority 61-CPR, root cost 61-RC5T, snd root link ID 61-RLID are set to match the link's priority 62-LPR, root cost 62-RCST, ~nd link ID 62-LID, with the system then returning to the block 91 to step on to the next linlc.
Thus the links are processed in sequence, ~s for st~ge I. When this pro-cess is complete, sta~e II is finished.
2~113~
St~Se III
St~ge III ~Fi~. ~;C~ follows. whlch a~ain processes the links in seau'ence.
In this ste~;e, e~ch link ls set to the HUD or HUE state and its Hello-messa~se priorlty is set. Block 100 is a decision blocL which causes the links listed in the link table 62 to be processed in sequence. For e9ch link, the Hello-messsge priority ~2-HMPR is set (block 101) to the highest of the priorities of all other llnks (excludin~ the llnk belng processed itself) ~nd the priorlty of the hybrid unit. Block 102 tests whether the link ID 62-LID of the link is the same ~s that in the root link ID register 61-RLID; if it is, then the link is set to HUE (block 103). If it is not, block 104 te-~ts whether the Designation 62-DES of the link is 1 (true). If it is, the state 62-ST is set to HUD (block 105); if it is not, the stste is set to a special null state HUN (block 106) in which the hybrid unit sends only HU-Hello mess~ges on that link, but no DU-Hello or EU-Hsllo messa~Ses.
When all links hsve been processed in st0~e III, the spanning tree al8o-rithm is finished.
.
Deletion clocks 68 and 69 operete in con~unction with timer sections 62-T and 63-T in tables 62 and 63 to remove entries from these t~bles if no messages ~re received to update them within suitable time limits. The operation of the spsn-ning tree ~l~sorithm applied ot each hybrid unit csn be described ~long with this circuitry.
HU-Hello messages received by an HU-Hello receiver 67 are used to maintain information about the neighbouring hybrid units ln the hybrid unit neighbour table 63. The inform~tion consists of: the identification of the link on whichthe nelghbour exists (link ID). held in e section ô3-LID; the priority value contalnéd in the HU-Hell~ messa~e, held in e section 63-PR; the cost value con-tained in the HU-Hello message (the reported cost of the p3th from the neigh-bour hybrid unit to the root hybrid unit), held in a section 63-CST; the identi-fier of the sendin~; hybrid unit contained in the HU-Hello mess~ge, held in a section 63-ID. ~On a multi-access llnk such as a LAN, there may of course be many neighbouring hybrid units).
The hybrid unit elso malntains a link t~ble 62. Each entry in this table records information 3bout one of the links of the hybrid unit. The link ID, in section 62-LID, is the link identifier (unique to the link within the hybrid unit~.
The cost, held in section 62-CST, is o v~lue set by the operrtor to determine a preference for choosin~s one link over ~nother in the spannin~ tree dgorithm.
The distribution unit nelghbour, held in section 62-DU, is o 13Ooleon value (s - 21 ~
sln~le blt) set by the DU-Hello receive lo$ic 44 of Fi8- 4 to I ~true~ if the distribution unit nei$hbour table 45 of Fi~. 4 contains an entry for a distrlbu-tion unit on that link. and to n (false! otherwise.
Each link entry in the link table 62 includes a link priority section, ô2-LPR. The m~nner in which the link priority i5 determine~ is described below.
Esch link entry also contains a Hello-messa~e priority, held in section 62-HMPR,which Is the hi~shest of the link priorities of all other links in the link tebla and the prlority of the hybrid unit itself. The HU-Hello generotor 66 periodi-cally sends, on each of the linkQ of the hybrid unit. HU-Hello me sagec, the contents of which are link-dependent; in particular, the priority field of ~n HU-Hello messe$e is set to the priority value 62-HMPR from the appropriste entry in the link table 62. and the cost field is set to the velue in the re~ister 61-RC~T.
The set of registers 61 include~ a hybrid unit identifier re~ister 61-KUID
and a hybrid unit prlority reE~ister ôl-HUPR. Register 61-HUID stores the identifier HU-ID of the hybrid unit. This must be unique emon~ all hybrid unlts in e subnetwork; it may be congi~sured by the operator or -be derived fromsome other unique property of the hybrid unit. Re~ister 6 l-HUPR stores the hybrid unit priority of the hybrid unit. This is set by the operator to deter-mine a preference for one hybrid unit in the subnetwork over others in choosinE~the spanning tree root.
The other three re~isters in the set of registers 61 are set by the control lo~ic 64, which elso sets values in the link table 62. The control logic 64 implements the spenning tree algorithm. which consists of three successive sta~ses. The operetion of the control logic 64 c~n be tri8~ered in response to eny chan~e in any o~ the values used AS inputs to the lo$ic from any of the table~.
~ efore the first executlon of stage I. the system is initialized by settin~the re$ister ôl-RCST to O and, in the link table, section 62-HMPR to the value <1, 61-HUPR, ôl-HUID> for each link.
.
2B7~30 St~e I
.
In sta~e I, the control logic t;4 determinec, for earh link, a priori~y, 3 root cost, snd a desi~nstlon, and sets these in the entry for that link in the link table 62 (in. the priority section 62-LPR, the root cost section 62-RCST, ~nd the design~tlon section 62-DES respectively). The Deslgn~tion field 62-DFS for each link is used on multi-cccess links to select one hybrid unit only to act ~sDU on th~t link in the case where more thon one hybrid unit m~ly compute the link to be ~wey from the sp~nnin~ tree root. The hybrid unit with highest priority on the link will set Desi~nation to l (true) and ~rt e i o DU, while all other hybrid units on the link, for which the link is flW~y from the spsnning tree root, will set Desi~n~tion to 0 (felse) and send ond receive only HU-Helloson th~t link.
For eech link in the t~ble 62, the control lo~ic finds the hi~hest priority from the set of priorities derived from the priorities of ~11 nei~hbours on the link ~obtsined from the Nei~hbo~lr teble 63), ~nd includinE~ the priority sent ln HU-Hellos on this link by the hybrid unit itself.
Fig. ôA is ~n informsl flow di~gram for stsge I. The first block, block 80, ls fl decision block which c~uses the links listed in the link table 62 to be processed in sequence. For each link, the appropri~te sequence from the blocks 81 to 86 is performed; control then returns to block 80, and the next link is processed. When ~11 the links h~ve been processed, sta~e I is complete, and the N exit from block 80 ls taken, to st~ge II (Fi~. ôB).
.
After a link h~s been selected in block 80, the first operations performed are those of block 8l, initifllizing the link volues for that link. Thus the link design~tion 62-DES is set to l (true), the link priority 62-LPR is set to the priority for the Hello mess~es 62-HMPR. end the link root cost 62-RCST is set to the root eost 61-RCST in register set 61.
The distribution unit number 62-DU for the llnk is ther, examined in block 82. If this is l, indicatin~ that the link hfls 9 distribution unit neighbour,then block 83 is performed. In this block, the link priority 62-PR is set to <0, 61-HUPR, ôl-HUID>. The root cost for the link, 62-RCST,is set to the cost for the link, 62-C3T. Fin~lly, the designatlon for the link 62 DES is set to 0.
2~7~1~0 If the distribution unit num~er 62-DU is found to be 0 in block 80, then block 84 follows. This deçision block causes the hybrid units listed in the nel~hbour tsble 6~ for th~ current link to b~ process~d in sequence: wheri 911 hsve been processed, the operation returns to block 80 for the next link.
Block 84 is followed by block 85, which tests whether the priority fleld <62-PR, 62-RCST, 61-HUID> ls ~re~ter than the priority field <63-PR, (63-CST +
62-CST), 63-ID>. If it is not, then no ~ctlon is tsken, snd the system returnsto block 84 ~nd ~dvences to the next hybrid unlt for the link. If it is, then in block 86, the link priority 62-PR i9 replaced by the hybrid unit priority 63-PR, the llnk root cost 62-RCST is set to the sum of the link co~t 62-CST of thls link and the root cost 63-CST reported in HU-Hellos from this neighbour, ~nd the link desi~nation 62-DES is set to 0; the system then returns to block 84snd advsnces to the next distribution unit for the link.
Thus the links are processed in sequence, and for each link with hybrid units ~s nei~hbours, the processin~ is for those hybrid unlts in sequence. As result of sts$e I, the control logic 64 sets the value of priority 62-PR, root cost 62-RCST, snd Desi~n~tion 62-DES for eoch link in the link t~ble. When this process is complete. st~e I is finished.
Sta~5e II
After the priorities, root costs, and desi~nctions of 911 the links hsve been determined in sta~e I, stoge II follnws, which c8~in pro~esses the links in sequence. In this st~e, the control logic 64 determines the highest priority link - i.e. the link with the hl~hest priority 62-PR ~nd root cost 62-RCSI'.
If more thcn one link hes the s~me highest priority, the link with the highest link ID (section ô2-LID of the link entry in t~ble 62) is chosen. The link ID
of the hiEShest priority circuit is set in the root link ID re$ister 61-RLID. Ifthe hybrid unit itself hss 3 hi~her priority th~n the priority of sny of its links, then the root link ID re~ister is set to 0. This me~ns th~t there is no root link, and the hybrid unit is the spsnnin~ tree root in ~n isol~ted subnet-work. The hl~shest priority is set in the cslculated priority ree;ister 6 l-CPR.The root cost for the hi~hest priority link (or zero if the HU is the spannin~
tree root snd h~s no DU nei~hbour) is set ln the re~ister root cost 61-RCST.
2-~:L1 3~
~ 24 --Sts~e II (Fi~. 6~3) storts with an initi91izing block 90 whlch initializes both the hybrid unit root link ID in re~ister 61-RLID snd the hybrid unit root cost in reo~ister 61-RCST to 0 This is followed by block 90. ~ decision block which c~uses the links listed in the link tcble 62 to be processed in sequence. For each link, the sppropri~te sequence from the blocks 92 to 98 is performed: con-trol then returns to block 90, ~nd the next link ic processed. When all the links h~ve been prccessed, st~ge II is complete, and the N exit from block 90 istsken, to stcge III (Fi$. 6C~.
Block 92 tests whether the link priority 62-PR is equ~l to the Hello-mess&ge priority 62-HMPR for HU-Hello mess~e~ sent on thst llnk. If it is, nothinES further is done, end the system returns to block 91 to step on to the next link. If it is not, the next block is 93.
Block 93 compares the priority field <61-CPR, 61-RC5T> of the hybrid unlt with the priority field <62-PR, 62-RCST> of the link. If the result ls U~ the hybrid unit's prlority field is less th~n the link's priorlty field,~ i then no action is t~ken, and the system returns to block 91 to step to the next link. If the two fields ~re equol, then checks Qre made to see whether the hybrid unit's link identifier 61-RLID ls O ~block 95) and whether the link's identifier 62-LID is less thon the hybrid unit's root link identifier 61-RLID
~block 96). If the hybrid unit's link identifier is O or is gre~ter than the link's identifier, then then no action is taken, ~nd the system returns to block91 to step to the next link. If the hybrid unit's link identifier is ~reDter th~n 0 aDd less than the link's identifier, then it it set to eqtlal the link's identifier (block 97), and tho system then returns to block 91 to step to the next link. Fln~lly, if the hybrid unit's priorlty fleld is ~Sre~ter th3n the link's priority field, then the hybrid unlt's cost priority 61-CPR, root cost 61-RC5T, snd root link ID 61-RLID are set to match the link's priority 62-LPR, root cost 62-RCST, ~nd link ID 62-LID, with the system then returning to the block 91 to step on to the next linlc.
Thus the links are processed in sequence, ~s for st~ge I. When this pro-cess is complete, sta~e II is finished.
2~113~
St~Se III
St~ge III ~Fi~. ~;C~ follows. whlch a~ain processes the links in seau'ence.
In this ste~;e, e~ch link ls set to the HUD or HUE state and its Hello-messa~se priorlty is set. Block 100 is a decision blocL which causes the links listed in the link table 62 to be processed in sequence. For e9ch link, the Hello-messsge priority ~2-HMPR is set (block 101) to the highest of the priorities of all other llnks (excludin~ the llnk belng processed itself) ~nd the priorlty of the hybrid unit. Block 102 tests whether the link ID 62-LID of the link is the same ~s that in the root link ID register 61-RLID; if it is, then the link is set to HUE (block 103). If it is not, block 104 te-~ts whether the Designation 62-DES of the link is 1 (true). If it is, the state 62-ST is set to HUD (block 105); if it is not, the stste is set to a special null state HUN (block 106) in which the hybrid unit sends only HU-Hello mess~ges on that link, but no DU-Hello or EU-Hsllo messa~Ses.
When all links hsve been processed in st0~e III, the spanning tree al8o-rithm is finished.
.
Claims (8)
- Claims In a communication network comprising end units (EUs) and distribution units (DUs) coupled together by communication links. each end unit having at least one network service access point (NSAP) or address which identifies the unit and each distribution unit including routing tables for routing data packets through the network between end units, the units maintaining knowledge of their neighbouring units by exchanging Hello messages which indicate the unit type and contain the NSAP(s) of the end units, the DUs collectively maintaining complete information about all NSAPs, a hybrid unit (HU) having one link (HUE
link) on which it simulates en EU and a plurality of links (HUD links) on which it simulates DUs, sending on the HUE link EU-Hello messages containing the NSAPs of all EUs which it receives on its HUD links. and sending on the HUD
links DU-Hello messages. - 2 A hybrid unit according to claim 1 wherein the communication links include local area networks (LANs).
- 3 A hybrid unit according to either claim 1 or 2 wherein there is a prede-termined limit on the number of NSAPs which the routing table of an EU can contain.
- 4 A hybrid unit according to claim 1 or 2 , wherein the routing table contained in the hybrid unit comprise an NSAP table which contains a list of all NSAPs contained in EU-Hello messages received by the HU, and a distribution unit table which contains a list of distribution units to which the hybrid unit is linked.
- 5 A hybrid unit according to claim 1 or 2 wherein a hybrid unit is reconfigurable to change the assignations of its links.
- 6 A hybrid unit according to claim 5 wherein the hybrid unit is manually reconfigurable.
- 7 A hybrid unit according to claim 5 wherein the hybrid unit is automatically reconfigurable.
- 8 A hybrid unit according to claim 7 wherein the hybrid units implement a spanning tree algorithm which selects a single hybrid unit from a connected sub-network of hybrid units as a root unit presenting HUD links to all neighbouring hybrid units, the selection being of a hybrid unit neighbouring a distribution unit if the subnetwork is not isolated.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| GB919112898A GB9112898D0 (en) | 1991-06-14 | 1991-06-14 | Communication networks |
| GB9112898.3 | 1991-06-14 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CA2071130A1 true CA2071130A1 (en) | 1992-12-15 |
Family
ID=10696715
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CA002071130A Abandoned CA2071130A1 (en) | 1991-06-14 | 1992-06-12 | Communication networks |
Country Status (6)
| Country | Link |
|---|---|
| US (1) | US5491692A (en) |
| EP (1) | EP0518581B1 (en) |
| JP (1) | JPH05199227A (en) |
| CA (1) | CA2071130A1 (en) |
| DE (1) | DE69207822T2 (en) |
| GB (1) | GB9112898D0 (en) |
Families Citing this family (62)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE69331705T2 (en) * | 1992-12-21 | 2002-12-19 | Apple Computer | METHOD AND DEVICE FOR CONVERTING ANY TOPOLOGY FROM A NODE COLLECTION IN AN ACYCLIC DIRECTED GRAPH |
| JP3425192B2 (en) * | 1993-08-25 | 2003-07-07 | 富士通株式会社 | Address information automatic setting processing method and address information setting device |
| SE503219C2 (en) * | 1994-09-05 | 1996-04-22 | Ericsson Telefon Ab L M | Device and process for process-based message management in a communication system |
| US20030157074A1 (en) * | 1994-11-16 | 2003-08-21 | Mitrani Eduardo N. | Vitro micro-organs, and uses related thereto |
| US5555478A (en) * | 1995-06-07 | 1996-09-10 | Zelikovitz, Deceased; Joseph | Fiber optic information transmission system |
| US5917820A (en) * | 1996-06-10 | 1999-06-29 | Cisco Technology, Inc. | Efficient packet forwarding arrangement for routing packets in an internetwork |
| US6757286B1 (en) | 1997-03-24 | 2004-06-29 | Alcatel | Self-configuring communication network |
| JP3112160B2 (en) * | 1997-09-29 | 2000-11-27 | 日本電気株式会社 | Automatic detection method of network address duplicate device |
| US6339595B1 (en) | 1997-12-23 | 2002-01-15 | Cisco Technology, Inc. | Peer-model support for virtual private networks with potentially overlapping addresses |
| US7369556B1 (en) | 1997-12-23 | 2008-05-06 | Cisco Technology, Inc. | Router for virtual private network employing tag switching |
| GB2332809A (en) * | 1997-12-24 | 1999-06-30 | Northern Telecom Ltd | Least cost routing |
| US7687057B2 (en) * | 1998-01-09 | 2010-03-30 | Yissum Research Development Company Of The Hebrew University Of Jerusalem | In vitro micro-organs, and uses related thereto |
| US5991300A (en) * | 1998-09-08 | 1999-11-23 | Cisco Technology, Inc. | Technique for efficiently performing optional TTL propagation during label imposition |
| US6295296B1 (en) | 1998-09-08 | 2001-09-25 | Cisco Technology, Inc. | Use of a single data structure for label forwarding and imposition |
| US6614781B1 (en) * | 1998-11-20 | 2003-09-02 | Level 3 Communications, Inc. | Voice over data telecommunications network architecture |
| EP1009130A1 (en) * | 1998-12-11 | 2000-06-14 | International Business Machines Corporation | Distributed directory services for locating network resources in a very large packet switching network |
| US7307990B2 (en) * | 1999-01-19 | 2007-12-11 | Cisco Technology, Inc. | Shared communications network employing virtual-private-network identifiers |
| US6337861B1 (en) | 1999-02-02 | 2002-01-08 | Cisco Technology, Inc. | Method and apparatus to properly route ICMP messages in a tag-switching network |
| US6512768B1 (en) | 1999-02-26 | 2003-01-28 | Cisco Technology, Inc. | Discovery and tag space identifiers in a tag distribution protocol (TDP) |
| US6473421B1 (en) | 1999-03-29 | 2002-10-29 | Cisco Technology, Inc. | Hierarchical label switching across multiple OSPF areas |
| GB9913239D0 (en) | 1999-06-08 | 1999-08-04 | Marconi Comm Ltd | Communications arrangement |
| US20030152562A1 (en) * | 2001-10-23 | 2003-08-14 | Yissum Research Development Company Of The Hebrew University Of Jerusalem | Vitro micro-organs, and uses related thereto |
| US20020110108A1 (en) * | 2000-12-07 | 2002-08-15 | Younglok Kim | Simple block space time transmit diversity using multiple spreading codes |
| US20030046378A1 (en) * | 2001-07-14 | 2003-03-06 | Zimmel Sheri L. | Apparatus and method for existing network configuration |
| US20030051007A1 (en) * | 2001-07-14 | 2003-03-13 | Zimmel Sheri L. | Apparatus and method for optimizing telecommunication network design using weighted span classification for low degree of separation demands |
| US20030023706A1 (en) * | 2001-07-14 | 2003-01-30 | Zimmel Sheri L. | Apparatus and method for optimizing telecommunications network design using weighted span classification and rerouting rings that fail to pass a cost therehold |
| US20030035379A1 (en) * | 2001-07-14 | 2003-02-20 | Zimmel Sheri L. | Apparatus and method for optimizing telecommunication network design using weighted span classification |
| WO2003040686A2 (en) * | 2001-11-05 | 2003-05-15 | Medgenics, Inc. | Dosing and administration of therapeutic micro-organs in living subjects and devices and methods for same |
| US7296061B2 (en) * | 2001-11-21 | 2007-11-13 | Blue Titan Software, Inc. | Distributed web services network architecture |
| US7853643B1 (en) | 2001-11-21 | 2010-12-14 | Blue Titan Software, Inc. | Web services-based computing resource lifecycle management |
| JP4326189B2 (en) * | 2002-06-10 | 2009-09-02 | 健 坂村 | Autonomous IC card and communication system |
| AU2003242967A1 (en) * | 2002-07-12 | 2004-02-02 | Yissum Research Development Company Of The Hebrew University Of Jerusalem | Method and device for inducing biological processes by micro-organs |
| JP4056864B2 (en) * | 2002-11-26 | 2008-03-05 | 富士通株式会社 | Relay device |
| US7508801B1 (en) | 2003-03-21 | 2009-03-24 | Cisco Systems, Inc. | Light-weight access point protocol |
| US7313113B1 (en) | 2003-04-04 | 2007-12-25 | Airespace, Inc. | Dynamic transmit power configuration system for wireless network environments |
| US7301926B1 (en) | 2003-04-04 | 2007-11-27 | Airespace, Inc. | Automatic coverage hole detection in computer network environments |
| US7340247B1 (en) * | 2003-05-29 | 2008-03-04 | Airespace, Inc. | Wireless network infrastructure including wireless discovery and communication mechanism |
| US7539169B1 (en) | 2003-06-30 | 2009-05-26 | Cisco Systems, Inc. | Directed association mechanism in wireless network environments |
| US7729267B2 (en) * | 2003-11-26 | 2010-06-01 | Cisco Technology, Inc. | Method and apparatus for analyzing a media path in a packet switched network |
| US20050201372A1 (en) * | 2004-02-11 | 2005-09-15 | Susan Hares | Systems and methods for virtual broadcast subnetworks |
| US7433696B2 (en) * | 2004-05-18 | 2008-10-07 | Cisco Systems, Inc. | Wireless node location mechanism featuring definition of search region to optimize location computation |
| US7516174B1 (en) | 2004-11-02 | 2009-04-07 | Cisco Systems, Inc. | Wireless network security mechanism including reverse network address translation |
| US7433362B2 (en) * | 2004-11-05 | 2008-10-07 | Sbc Knowledge Ventures, L.P. | System and method for SONET NSAP addressing |
| US7457262B1 (en) | 2004-11-05 | 2008-11-25 | Cisco Systems, Inc. | Graphical display of status information in a wireless network management system |
| US7596376B2 (en) * | 2005-02-18 | 2009-09-29 | Cisco Technology, Inc. | Methods, apparatuses and systems facilitating client handoffs in wireless network systems |
| US7805140B2 (en) * | 2005-02-18 | 2010-09-28 | Cisco Technology, Inc. | Pre-emptive roaming mechanism allowing for enhanced QoS in wireless network environments |
| US8559443B2 (en) | 2005-07-22 | 2013-10-15 | Marvell International Ltd. | Efficient message switching in a switching apparatus |
| US7852805B2 (en) * | 2005-11-01 | 2010-12-14 | Kahtava Jussi T | Variable length radio link ID for resource allocation in mobile communication systems |
| US7821986B2 (en) * | 2006-05-31 | 2010-10-26 | Cisco Technology, Inc. | WLAN infrastructure provided directions and roaming |
| US7499718B2 (en) | 2006-08-01 | 2009-03-03 | Cisco Technology, Inc. | Enhanced coverage hole detection in wireless networks |
| AU2007294858B2 (en) * | 2006-09-14 | 2011-05-12 | Medgenics Medical Israel, Ltd | Long lasting drug formulations |
| US7738383B2 (en) * | 2006-12-21 | 2010-06-15 | Cisco Technology, Inc. | Traceroute using address request messages |
| US7706278B2 (en) * | 2007-01-24 | 2010-04-27 | Cisco Technology, Inc. | Triggering flow analysis at intermediary devices |
| US7596461B2 (en) * | 2007-07-06 | 2009-09-29 | Cisco Technology, Inc. | Measurement of air quality in wireless networks |
| KR101591887B1 (en) | 2010-06-15 | 2016-02-04 | 메드제닉스 메디칼 이스라엘 리미티드 | Long lasting drug formulations |
| US8774010B2 (en) | 2010-11-02 | 2014-07-08 | Cisco Technology, Inc. | System and method for providing proactive fault monitoring in a network environment |
| US8559341B2 (en) | 2010-11-08 | 2013-10-15 | Cisco Technology, Inc. | System and method for providing a loop free topology in a network environment |
| US8982733B2 (en) | 2011-03-04 | 2015-03-17 | Cisco Technology, Inc. | System and method for managing topology changes in a network environment |
| US8670326B1 (en) | 2011-03-31 | 2014-03-11 | Cisco Technology, Inc. | System and method for probing multiple paths in a network environment |
| US8724517B1 (en) | 2011-06-02 | 2014-05-13 | Cisco Technology, Inc. | System and method for managing network traffic disruption |
| US8830875B1 (en) | 2011-06-15 | 2014-09-09 | Cisco Technology, Inc. | System and method for providing a loop free topology in a network environment |
| US9450846B1 (en) | 2012-10-17 | 2016-09-20 | Cisco Technology, Inc. | System and method for tracking packets in a network environment |
Family Cites Families (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US4984235A (en) * | 1987-04-27 | 1991-01-08 | Thinking Machines Corporation | Method and apparatus for routing message packets and recording the roofing sequence |
| US4905233A (en) * | 1987-11-23 | 1990-02-27 | Harris Corporation | Multiple path routing mechanism for packet communications network |
| CA1294347C (en) * | 1988-05-05 | 1992-01-14 | Man Him Hui | Remote interconnection of local area networks |
| US4864559A (en) * | 1988-09-27 | 1989-09-05 | Digital Equipment Corporation | Method of multicast message distribution |
| US5060228A (en) * | 1988-11-19 | 1991-10-22 | Fujitsu Limited | Bridge communication system |
| JPH02182057A (en) * | 1989-01-09 | 1990-07-16 | Canon Inc | Network management system |
| US5309437A (en) * | 1990-06-29 | 1994-05-03 | Digital Equipment Corporation | Bridge-like internet protocol router |
| US5179555A (en) * | 1990-09-11 | 1993-01-12 | Microcom Systems, Inc. | High speed data compression and transmission for wide area network connections in LAN/bridging applications |
| US5159592A (en) * | 1990-10-29 | 1992-10-27 | International Business Machines Corporation | Network address management for a wired network supporting wireless communication to a plurality of mobile users |
| US5265092A (en) * | 1992-03-18 | 1993-11-23 | Digital Equipment Corporation | Synchronization mechanism for link state packet routing |
-
1991
- 1991-06-14 GB GB919112898A patent/GB9112898D0/en active Pending
-
1992
- 1992-06-08 DE DE69207822T patent/DE69207822T2/en not_active Expired - Fee Related
- 1992-06-08 EP EP92305212A patent/EP0518581B1/en not_active Expired - Lifetime
- 1992-06-11 JP JP4152104A patent/JPH05199227A/en active Pending
- 1992-06-12 CA CA002071130A patent/CA2071130A1/en not_active Abandoned
-
1993
- 1993-12-23 US US08/173,719 patent/US5491692A/en not_active Expired - Lifetime
Also Published As
| Publication number | Publication date |
|---|---|
| EP0518581B1 (en) | 1996-01-24 |
| JPH05199227A (en) | 1993-08-06 |
| US5491692A (en) | 1996-02-13 |
| EP0518581A1 (en) | 1992-12-16 |
| DE69207822D1 (en) | 1996-03-07 |
| GB9112898D0 (en) | 1991-07-31 |
| DE69207822T2 (en) | 1996-09-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CA2071130A1 (en) | Communication networks | |
| US5764636A (en) | Color blocking logic mechanism for a high-performance network switch | |
| JP3662378B2 (en) | Network repeater | |
| TWI813742B (en) | Asynchronous object manager in a network routing environment | |
| CN100444563C (en) | Exchange node classifying and marking rules | |
| CN101330370B (en) | Node and communication method | |
| CN100574242C (en) | The method and apparatus of relaying in switch | |
| CN101141241B (en) | Method and network appliance for implementing MAC safety | |
| CN100377529C (en) | Route designing method | |
| US7532588B2 (en) | Network system, spanning tree configuration method and configuration program, and spanning tree configuration node | |
| JP3748216B2 (en) | Packet transfer method and packet transmitter / receiver | |
| US6526052B1 (en) | Virtual local area networks having rules of precedence | |
| US5557745A (en) | Method for supporting foreign protocols across backbone network by combining and transmitting list of destinations that support second protocol in first and second areas to the third area | |
| JP2002314571A5 (en) | ||
| US7016979B2 (en) | System and method of accessing and transmitting different data frames in a digital transmission network | |
| US7522589B2 (en) | Network units for use in and organisation of cascade systems | |
| JP4701152B2 (en) | Data relay apparatus, data relay method, and data relay program | |
| JP2005517353A (en) | Sending packets over aggregated links using ingress card distributed processing | |
| CN104272679A (en) | Communication system, control device, communication method, and program | |
| CN103026729B (en) | For key-course and the method for multi-stage light burst switching system | |
| HU216230B (en) | Method for selecting links between a source and a destination in a transport network | |
| CN101911602A (en) | The method and system that comprises the telecommunication of the self-organizing scalable Ethernet that uses the IS-IS level | |
| CN100502329C (en) | Ethernet exchanger and its service processing method | |
| US20150033010A1 (en) | Method for the secure exchange of data over an ad-hoc network implementing an xcast broadcasting service and associated node | |
| JP2001127785A (en) | Network connection device, communication unit and network connection method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| EEER | Examination request | ||
| FZDE | Discontinued |