GB2280339A - Wide area bridging path selection - Google Patents

Abstract

In a method and apparatus for the interconnection of local area networks by transparent bridging, two networks LAN1, LAN2 are coupled by a bridge pair B1, B2 which include a plurality of alternative communication links L1, L2, L3 between them. Data frames are transmitted from network to network over one the links, the link being selected according to a particular function applied to the unique source address of the device D1 - D6 sending the data frame. The function is chosen to optimise the distribution of data frames over the different links by appropriate selection of the number and position of the source address bits used as input to the selection function. Transparency of the bridging operation is maintained, and there is no conflict with spanning tree protocols. <IMAGE>

Description

WIDE AREA BRIDGING PATH SELECTION The present invention relates to a method and apparatus for interconnecting local area networks (LANs) by way of a pair of communicating wide area bridging devices, and in particular to a method and apparatus for selecting a path for the transmission of data frames between the bridging devices where there are a plurality of possible communication paths therebetween.

It is now commonplace to connect together multiple local area networks, each of which has coupled thereto a plurality of data processing devices, in order to extend the coverage of the individual LANs. Various techniques and protocols exist for the coupling of LANs to form wide area networks (WANs), and the present invention relates particularly to the use of transparent bridges in the implementation thereof.

In such systems, the transfer of data frames from a particular source device on one LAN to a destination device on another LAN takes place across such a bridge without any visibility of this transfer by either the source or destination devices. From the point of view of the source and destination devices, they essentially share the same LAN, and the bridging operation is said to be transparent.

In the implementation of such bridging techniques, a protocol must exist which prevents the duplication of data frames between devices, and which prevents the reseqencing of any successive data frames between two devices.

This is commonly achieved by the network defining a spanning tree which ensures that no loops are permitted in the network. Such techniques are well known, and described, for example in the art.

There are, however, drawbacks to defining a single spanning tree which permits communication between the various LANs using only single paths, in that the use of redundant links can be advantageous in reducing congestion on the networks, or providing additional reliability of the overall network by mitigating the loss of single communication links. Some systems use multiple spanning trees to effect this built-in redundancy, allowing use of all links between LANs. Which spanning tree to use for a given data frame transfer between devices may be determined by the source device, either explicitly or implicitly, for example using a logical combination of source and destination addresses.

Other techniques for the utilisation of redundant links are also known, but involve fairly complex protocols to permit the coupling of an entire network of wide area bridges including the use of such multiple paths between different bridge pairs. Various distributed load sharing paths are identified by the system as alternatives to the paths defined by the spanning tree, and only used under particular circumstances determined according to the relative positions of the cooperating bridges to the root bridge of the entire network.

Such techniques as described, however, are primarily designed for managing redundant links in a network topology such as that illustrated in figure 1, where, for example, local area networks LAN1...LAN4 are mutually connected using bridge devices B1...B4 and links LSy, where x and y indicate the respective LAN or bridge numbers in the connection. It can clearly be seen that a data frame originating from a device on LAN1 may be transmitted to a device on LAN4 by three possible paths: L12-L24; L14; and L13-L34, resulting in multiplication of data frames, or possible resequencing of frames arriving at LAN4 which have been sent via different paths.A spanning tree protocol such as that described above can be used to ensure that only one path may be used by any source-destination device pair, and that any other paths may be identified as redundant links.

The present invention is particularly relevant, in its simplest form, to a network topology in which multiple parallel communication paths are available between any two bridge devices, such as the configuration shown in figure 2.

Typically, two bridge devices B1 and B2, each connected to its respective LAN, LAN1 and LAN2, may be coupled by a plurality of communication paths L1, L2, L3 which may be provided to improve the reliability and/or capacity of the communication link or, for example, may be necessary where each link is of substantially lower transmission speed or bandwidth than the LANs being served. This situation is typically encountered where LANs are connected to form a WAN using multiple leased telecommunication lines, each of relatively low bandwidth.

It will be understood, however, that such parallel paths may be provided between bridge device pairs which themselves form part of network exemplified by figure 1 with alternative links (eg. L12--L24 vs. L13-L34) each including links with parallel paths. The systems and protocols previously referred to for managing use of redundant links are generally useable for managing parallel paths as well as the alternative (redundant) links, but are over-complex for this particular situation.

In such an arrangement as that shown in figure 2, since there is no guarantee of the data frame transfer time from LAN to LAN where data frames may be transmitted over any of the different links at different speeds, the frames may consequently be processed by the bridges in uncertain sequence.

Such potential resequencing of data frames is undesirable and must be avoided.

It is therefore necessary to provide a method and apparatus to ensure that data frames passing between the same two devices, eg. D1 and D5, always arrive at their destination in the same sequence in which they were sent. This can be effected by ensuring that such data frames always follow the same path, although this constraint is relaxed to not include data frames passing between different combinations of source and destination devices, eg. D2 and D5, or D2 and D6.

It is also desirable to maintain the transparency of the bridge to the source and destination devices; ie. they must not be required to play an active part in the choice of path used.

The present invention is directed toward a simple and effective way of providing a method and apparatus for the effecting of such data frame transfers between two participating bridges having multiple parallel communication paths therebetween, which guarantees that no resequencing of data frames between a particular source I destination device pair will take place. The present invention also provides well-distributed use of the available paths or links between the two bridges. The present invention may also provide the capability of accommodating differing speed paths between the bridges while maintaining such well-distributed use.

In accordance with one embodiment of the present invention, there is provided a bridge unit for interconnecting two local area networks. Each local area network includes a plurality of devices coupled thereto and each device has a unique address identifying said device. The bridge unit is coupled to a first LAN and has a plurality of output paths for forwarding data frames from a source device on said first LAN to a corresponding bridge unit coupled to a second LAN. The bridge unit comprises path selection means for selecting a one of said plurality of output paths for transmission of a data frame from said first LAN to said corresponding bridge unit, said selection means deriving said output path selection from an address contained within the data frame identifying the source device.

The present invention will now be described in detail, by way of example, and with reference to the accompanying drawings in which: Figure 1 shows an exemplary prior art bridging configuration for the coupling of a plurality of LANs; Figure 2 shows a bridging configuration for a pair of LANs suitable for the method and apparatus of the present invention; Figure 3 shows a LAN configuration and bridge unit according to the present invention; Figure 4 shows a flowchart illustrating a method according to one embodiment of the present invention.

With reference to figure 3, there is shown a pair of LANs (LAN1 and LAN2, coupled together by means of a bridge pair B1 and B2. LAN1 includes a number of devices D1, D2 and D3, and a connection 10 to bridge unit B1.

LAN2 includes a number of devices D4, D5 and D6, and a connection 20 to bridge unit B2. Bridge B1 listens to all data frame traffic placed onto LAN1 by any of the plurality of individual devices D1...D3, and forwards these data frames over a selected one of the links or paths L1...L3 to bridge B2, which subsequently places the data frames onto LAN2 for reception by any of devices D4..D6 (or even by a further bridge, not shown, for onward transmission to another LAN). The number of links is, for illustrative purposes, shown as three, but any number of suitable links may be provided dependent upon the traffic density of the networks. Although bridge B2 is not shown in detail, it will be understood that for traffic originating from LAN2, bridge B2 will operate in analogous fashion to that which is now described in respect of B1.

Any data frames originating from, for example, the device D1 and directed to a specific device or group of devices, for example D5, are, in accordance with the present invention directed by the bridge B1 over the same path Ln, where n may be any one of links L1...L3. Data frames directed to a plurality of destination devices must be handled in like manner. Similarly, any data frames originating from device D2 and directed to, for example, D6 must be transmitted over the same path, although this path may, of course be different from Ln above.

The data frames transmitted by any source device to any destination device always follow a predefined format which includes a source address uniquely identifying the source device, and a destination address uniquely identifying the destination address. Such addresses are known as media access control (MAC) addresses. At this level of protocol for bridging across networks, no part of the data frame specifies a route by which the frame must travel (a consequence of the transparency of the bridging operation).

Upon receiving a frame from the LAN1 (and assuming that the bridge B1 recognizes that the data frame is a frame which the bridge must forward), the bridging device B1 applies a hashing function to the source address contained therein in order to determine which of the possible paths L1..L3 is to be used to transmit the frame to bridge B2. Only the source address is used as input to the hashing function.

The function is chosen to be useful for a particular LAN protocol known as "Local Area Transport" (LAT) which is used in many computer systems including those built by Digital Equipment Corporation. A particular characteristic of systems which are using LAT is that they will normally also be using the DECnet protocol. This requires the devices on the networks to change their media access control address to have one of a particular range of values which is related to their DECnet address.

A DECnet format MAC address comprises a forty-eight-bit data sequence in the format (which is represented here in hex form): AA-00-04-OO-yz-wx16.

The first twenty-four bit sequence (eg. AA-00-0416) is a manufacturer code which will be common to all devices on the network which have the same manufacturer, and thus, as will become apparent is not a desirable sequence to use for the purposes of the present invention. The second twenty-four bit sequence (00-yz-wx,6) is a device address which is derived from a decimal format address of a.bl0, where a = area number, and b = node (device) number within the area.

The sequence wxyz,6 = (400,6 x a16) + b,6.

Thus, for the device with DECnet address 2.610, wxyz,6 = (40016 X 216) + 6l6 = 080616. The full MAC address is thus AA-00-04-00-06-08l6.

The inventor has recognized that the address of each device on a local area network, or wide area network, is unique and that this address may be used to determine which of a number of links (L1...L3) may be used to transfer data frames from bridge B1 to B2.

Furthermore, the inventor has recognized that in any network, it is only necessary to use a small portion of the source address to provide an efficient link determination mechanism for a given bridge, and that certain bits of the source address fields which may be regarded as "lowest order bits" are best suited to this application to optimise the even distribution of data frames over the available links.

In the present example of a DECnet protocol, it can be ascertained that the portion of the source device address which varies the most within any given LAN must be the lower order bits of the yz field, on the basis that the device addresses are assigned from 1 upwards.

A preferred hashing function according to the present invention selects the path Ln according to the value of an appropriately sized bit-field i of m bits, starting at the least significant bit of the yz octet, and working toward the most significant bit of the yz octet.

An exemplary function, according to the present invention, is given for a bridge supporting n parallel links (ie. Ln = Ll...Ln). The size, m, of the bit-field is chosen such that: 2 2 n > 2 For example, where the number of parallel links n is 2, the bit-field size m is 1; where the number of parallel links n is either 3 or 4, the bit-field size m is 2; where the number of parallel links n is from 5 to 8 inclusive, the bitfield size m is 3; where the number of parallel links n is from 9 to 16 inclusive, the bit field size m is 4. The bit-field i of size m is used as input to the hashing function to establish Ln where: n = (i mod n) + 1.

For five links Ll . . .L5 (n = 5), the bridge will ascertain the value of the final three bits of the hex field z. For: z = 0000; i = 000; n = 1 and link L1 is used; z = 0001; i = 001; n = 2 and link L2 is used; z = 0010; i = 010; n = 3 and link L3 is used; z = 0011; i = 011; n = 4 and link L4 is used; z = 0100; i = 100; n = 5 and link L5 is used; z = 0101; i = 101; n = 1 and link L1 is used; z = 0110; i = 110; n = 2 and link L2 is used; z = 0111; i = 111; n = 3 and link L3 is used.

(The first bit of hex field z is irrelevant for this i and could also be value 1.) In alternative address protocols to that described above, it will be necessary to ascertain which sub-fields of the MAC address vary most for each given LAN within the WAN. It will be understood that the hashing function need only be specific to each bridge, and can be chosen to optimise the distribution of device addresses for the particular LAN to which the bridge is attached. Frames transferred via the LAN from other parts of the network for onward transmission may have different values for the area portion of the address, but will also have a good distribution of device addresses. It will also be recognized that the path selection is entirely transparent to the source device, which need take no part in the path selection.

With further reference to figure 3, there is shown a bridge B1 according to the present invention. Data frames transmitted on LAN1 are received by bridge B1 on connection 10, and forwarded over one of three possible links L1...L3. The bridge B1 includes a source address identification unit 12 which is operative to locate and read the bit field i of the source address field forming part of the incoming data frame transmitted. This data is applied to path selection unit 14 to perform the hashing operation, using the output therefrom to control multiplexer 16. Multiplexer 16 diverts the data frame being transmitted over the appropriate link L1, L2 or L3.It will be readily apparent to one skilled in the art that a buffer arrangement (ordinarily provided, and not shown) is desirable to permit the processing of the source address by the bridge before actual onward transmission of the data frame commences. A hardware configuration table 18 may be maintained by the bridge to allow the path selection unit 14 to determine the availability of links for transmission. Such a table could be implemented in hardware, or could be a dynamic, software implemented system which is updated automatically when links fail or are added to the system. The path selection unit 14 can thereby automatically modify the hashing algorithm accordingly. In this event, a suitable mechanism must also be applied to prevent resequencing of data frames following a reconfiguration event.This could be implemented using a timing mechanism preventing transmission of a data frame for a predetermined period following an updating of the configuration table 18.

Also with reference to figure 4, there is shown a flowchart indicating the processing steps carried out by bridge B1 of figure 3. Each time a data frame is received on line 10, the number of output links currently available is determined by the path selection unit 14 by reference to the hardware configuration table 18, or other suitable method (step 100). In the event that any link should fail, the configuration table 18 is modified accordingly.

The path selection unit 14 then computes the number of bits required to form the input to the selection function (step 102). The appropriate bits of the data frame source address identification field are read by source address identification unit 12 according to the number of bits required (step 104) to form the input to the selection function.

The appropriate hashing operation is applied to the input function by path selection unit 14 to derive an output selection signal controlling multiplexer 16 to activate the appropriate link (step 106). The data frame is then transmitted over the selected link (step 108).

In the embodiments described above, the data frame traffic is distributed reasonably evenly (optimally where the number of lines Ln is a power of two), but without regard to the bandwidth of each communication path Ln. In a further refinement of the present invention, it is possible to accommodate variations in the path speeds by modifying the distribution of the input addresses i such that the hashing function maps more addresses i to links having higher bandwidth. This can be readily achieved by, for example, identifying (eg. in the hardware configuration table) a single physical link Ln with a number of link identifiers (Ln', Ln", Ln"'...) proportional to the bandwidth.

For example, a physical link Ln which has three times the bandwidth of the slowest link L1 can be identified as all three of links L2, L3 and L4. Any data frame hashed to link L2, L3 or L4 will be switched by path selection unit 14 and multiplexer 16 to the same physical link. Thus, three times the number of data frames (assuming an even distribution of source addresses) will be transmitted on this link. An alternative to this technique would be to use a simple look-up table.

Although the present invention has been described by way of examples showing a single direction transfer from a bridge B1 to a bridge B2, it will be understood that the bridge unit B2 is exactly analogous to bridge B1, and applies the hashing function to the address of devices D4, D5 and D6 in LAN2 for the transmission of data frames to devices on LAN1.

Although the present invention has been described with reference to a bridge unit interfaced with a single LAN, it will be understood that the invention is also applicable to a bridge unit which interfaces several LANs.

Claims (13)

1. A bridge unit for interconnecting two local area networks, each local area network including a plurality of devices coupled thereto and each device having a unique address identifying said device, the bridge unit having means for being coupled to a first LAN and having a plurality of output paths for forwarding data frames from a source device on said first LAN to a corresponding bridge unit coupled to a second LAN, said bridge unit comprising: path selection means for selecting a one of said plurality of output paths for transmission of a data frame from said first LAN to said corresponding bridge unit; said selection means including means for deriving said output path selection from an address contained within the data frame identifying the source device.

2. A bridge unit according to claim 1 wherein each device is uniquely identified by an N-bit address and wherein said path selection means includes means to identify and read a predetermined number of least significant bits from the address of said source device.

3. A bridge unit according to claim 2 wherein N = 48.

4. A bridge unit according to claim 3 wherein said address conforms to the DECnet media access control address protocol, and wherein said predetermined number of bits are selected from bit numbers 33-40 of the 48-bit DECnet MAC address.

5. A bridge unit according to claim 4 wherein said bridge unit is coupled to said corresponding bridge unit by means of n parallel paths, Ln, and said predetermined number of bits, m, is determined according to the expression: 2m 2 n > 2(=-l)

6. A bridge unit according to claim 5 wherein said path selection unit includes means to determine a selected path Ln according to the expression: n = (i mod n) + 1, where i is a number corresponding to the value of the m predetermined number of bits of the source device address.

7. A bridge unit according to any one of claims 1 to 5, wherein said path selection unit includes means to determine a weighting value to each path according to the bandwidth of said path.

8. A bridge unit according to any preceding claim including means to determine the number of output paths currently available.

9. A method of operating a bridge unit for interconnecting two local area networks, each local area network including a plurality of devices coupled thereto and each device having a unique address identifying said device, the bridge unit having means for being coupled to a first LAN and having a plurality of output paths for forwarding data frames from a source device on said first LAN to a corresponding bridge unit coupled to a second LAN, said method comprising the step of: selecting one of said plurality of output paths for transmission of a data frame from said first LAN to said corresponding bridge unit by deriving an output path selection code from an address, identifying the source device, contained within the data frame.

10. A method according to claim 9 wherein said source device address is an N-bit address and wherein said selection step includes reading a predetermined number of least significant bits from said address.

11. A method according to claim 10 wherein N = 48 and said source address conforms to the DECnet media access control address protocol, and wherein said selection step includes using a predetermined number of bits chosen from only bit numbers 33-40 of the 48-bit DECnet address.

12. A method according to claim 11 wherein there are n parallel output paths, Ln, and said predetermined number of bits, m, is determined according to the expression: 2m > n > 2(=-')

13. A method according to claim 12 wherein said output path Ln is selected according to the expression: n = (i mod n) + 1, where i is a number corresponding to the value of the m predetermined number of bits of the source device address.