System for providing fair access for VLANs to a shared transmission medium

[0002] Traditionally, many local area network (LAN) protocols indiscriminately send traffic to many or all other stations connected to the local area network. The addressing of many stations with a single message is generally called a ‘multicast’ communication. A special and frequent case of this is a message which has to be received by all stations connected to the network. Generally, this does not cause difficulty, but there may be occasions when the level of multicast traffic has limited the performance of the local area network. One type of occasion arises from the use of multicast messages for administrative purposes, for example to locate other stations on the network. When the number of stations on the same local area network becomes large so can the level of administrative multicasts. This may use significant proportions of the available bandwidth, i.e. the information carrying capacity of the network or may cause degradation of the end-stations' performance because a large number of these messages have to be processed.

[0003] There may be conditions wherein ‘rogue’ stations enter states whereby they continually transmit messages to the broadcast address. In theory, if such a condition were to arise, a large proportion of a network's band width could be utilised by these messages.

[0004] Further, if two network ports on the same local area network are connected together, a loop is produced. Ideally, the network devices would detect such a loop and shut down one of the relevant links but such a fault is not always corrected in this way. If a loop does occur in the local area network then the network is liable to become unusable.

[0005] More recently, networks have been designed to limit the effect of any ‘multicast’ storms that may happen. A common and preferred mechanism is to partition a local area network into a multiplicity of smaller local area networks, which are connected together using routers. Routers may monitor traffic and thereby enable a significant level of control over the passage of information between local area networks. Thus a ‘multicast storm’ on one local area network would not cause cessation of operation of another local area network.

[0006] If a system of this nature is employed, when a user wishes to connect a device to a specific local area network, the device is physically ‘patched’ through to a physical local area network implemented by a set of network devices. The only interaction between the local area networks is controlled by a router.

[0007] As networks have evolved, and as organisations have evolved, the use of hardwired networks using physical patch panels has become cumbersome. In simple terms, if physical patch panels are used then if a particular person in a organisation has shifted physical location, the physical change in location has to be accompanied by a physical change in the network topology via a patch panel. Additional network devices may be needed if changes in physical locations of users cause local network connections to be insufficient. Further, as organisations have split into smaller working units, the number of local area networks has expanded.

[0008] The concept of a virtual local area network allows a virtual patch panel to be implemented in the same devices that implement the network infrastructure. Each end station can be attached to a separate virtual local area network under software control. Virtual local area networks are, from the point of view of users or end stations, functionally similar to physical local area networks but are defined by software control of the addressing of data packets and other necessary or consequential control functions. For example, if a member (typically called client or user) in a virtual local area network has a message which is to be broadcast to all the members of the virtual local area network, then the message is treated as a multicast message and (by means well enough known not to require further description here) sent to all the other members of that virtual local area network but not to other users in the physical local area network which is partitioned into the multiplicity of virtual local area networks.

[0009] Nevertheless, somewhere in the network there must be a router which allows communication between the various virtual local area networks. Generally, each VLAN might need a separate physical connection to a router. This is inconvenient, because although the VLAN allows automatic connection of an end station to a VLAN, the network administrator must add an extra physical connection.

[0010] To avoid this, it is possible to multiplex multiple virtual local area networks over a single physical medium. A variety of schemes of achieving this exist but each depends on the tagging of each packet or data frame with a value identifying the VLAN to which the frame belongs.

[0011] This solves the problem of physical connections to the router but creates a new problem. The network was divided into VLANs to prevent traffic from one local area network from inhibiting access to all other networks. Now all the traffic from a multiplicity of VLANs is carried across a single physical connection, which has necessarily limited bandwidth. Again, a multicast storm on one segment may occupy all the available bandwidth on the shared, tagged link and again prevent or degrade performance on an unrelated local area network.

[0012] The invention is intended for use where data from a multiplicity of users in a plurality of virtual local area networks (VLANs) has to be multiplexed over a single physical link, and where, as indicated above, the aggregate traffic of all the VLANs together may exceed the maximum information carrying capacity of that physical link. The invention provides a scheme for according fair access to the single physical link.

[0013] A preferred but not in all respects essential technique according to the invention is as follows.

[0014] Means may be provided to monitor the traffic provided by each virtual local area network coupled to the physical link. The monitoring of that traffic may be gauged against that threshold or quota such that if the traffic from a particular VLAN exceeds a quota allocated thereto a state denoted ‘restricted’ is indicated in any desired manner. For a traffic flow below the quota, a ‘priority’ state is indicated. When the VLAN wishes to send a packet over the physical link it asserts a request signal (R). When the shared link is ready to accept a packet it may first search for requests from all VLANs which are indicated to be in the ‘priority’ state. A packet from a VLAN in the ‘priority’ state will always be accepted in preference to a packet from a VLAN in the ‘restricted’ state. If a VLAN in a ‘priority’ state sends a frame, that increment of traffic can cause the VLAN to exceed its quota, so that it will then enter the ‘restricted’ state. A variety of developments of such a scheme are possible. In particular, the quotas may be programmable or adaptive. They may for example be set in proportion to the number of ‘clients’ or users in each respective VLAN. Nevertheless the monitoring may indicate that the quotas for some VLANs need to be adjusted in accordance with monitor traffic flows to give greater preference to some VLANs over others. Furthermore, although the arbitration scheme envisaged as preferential is a round robin arbitration scheme so that a VLAN having priority is given unconditional access, it may be feasible to adapt the technique to a contention-resolution scheme, e.g. carrier sense multiple access, wherein a VLAN having priority is favoured in the contention-resolution process and a ‘restricted’ VLAN is not.

[0015] Various other schemes for setting the quota for the various VLANs connected to the physical link are possible. It is feasible to over subscribe the physical link. For example, if there were four ‘clients’ using a link, then the threshold for all the links could be set to 60%. The effect of this is that no single VLAN could consume more than 60% of the available bandwidth (information carrying capacity) if there were other VLANs which wish to send data.

[0016] FIG. 1 is a simplified diagram of a monitoring function performed on each VLAN connected to a shared transmission medium; and

[0017] FIG. 2 illustrates a system of control according to the invention.

[0018] FIG. 1 illustrates in simplified form part of a local area network wherein three virtual local area networks defined in the local area network share a common physical data transmission link. As indicated in the foregoing, a virtual local area network (VLAN) has all the benefits of a normal LAN but allows an administrator to configure which devices are connected to which VLAN through software control. This introduces a problem that the switch to which the device is to be attached must actually have a connection to each VLAN. Traditionally, this would have required a physical link between each switch for each VLAN to be supported. Such an arrangement does not scale. A solution is to allow traffic from a multiplicity of VLANs to be carried over a single link 1. Thus in the scheme of FIG. 1, packets from each of the three VLANs VLAN 1, VLAN 2 and VLAN3 are, under software control, tagged to identify their VLAN. A disadvantage to this scheme is that it is partially in contradiction to the idea of having separate VLANs. One of the reasons for partitioning the network into VLANs is to prevent excessive traffic from one group of users having an adverse effect on the service or performance of users on another VLAN. In a scheme such as FIG. 1, where traffic from a multiplicity of VLANs is carried across a single connection, typically an Ethernet connection, large bursts of traffic from one VLAN may saturate the link and prevent access to the medium from another VLAN.

[0019] Methods exist for the monitoring of traffic by examination of, for example, packet or frame headers including identifications of source, destination or other data. In the scheme illustrated in FIG. 1, the link (or port) 1 may be adapted to monitor all traffic from each VLAN independently. The monitoring is illustrated by the three notional ‘gauges’ 11, 12 and 13 for the VLANs VLAN 1, VLAN 2 and VLAN 3 respectively. Each of the gauges is shown with a threshold, denoted by the transition from shaded to clear portions in the gauge. If the measured traffic for a particular VLAN is above the respective threshold then that VLAN is deemed to have ‘restricted’ access to the physical medium. If the measured traffic for that VLAN is below the threshold, the VLAN has ‘priority’ access to the physical medium. Various conventional measures of traffic flow may be employed, such as bits (or octets) per unit time.

[0020] In the illustration of FIG. 1, VLAN 1 and VLAN 3 are providing the traffic above the threshold or quota and accordingly they are deemed to have restricted access to the physical link, whereas the traffic from VLAN 2 is below the quota and accordingly VLAN 2 is deemed to have priority access to the physical link.

[0021] In the simple scheme, a frame from VLAN 2 can always be accepted in preference to a frame from the other VLANs. Such a condition would continue until VLAN 2 exceeded its quota or one of the other VLANs had sent so little data that it entered the priority state.

[0022] One monitoring mechanism is to provide a traffic counter which is decremented at a fixed rate and incremented each time traffic is sent from the respective VLAN to the link. A threshold is set according to the relevant averaging period to implement the required quota. Generally, when the average data rate from a VLAN is below its quota the counter will decrement more quickly than it increments and so it will be permanently ‘empty’, i.e. at a state which denotes emptiness. If the data rate exceeds the decrementing rate then the counter will gradually increment until it exceeds a predetermined count which denotes the quota. Such a monitoring mechanism is conveniently called a ‘leaky bucket’ and is more particularly described in the commonly assigned co-pending application Ser. No. 09/093,287 (based on GB application number 9807264.8 filed Apr. 3, 1998).

[0023] FIG. 2 illustrates how the control may be achieved. As in FIG. 1, the VLANs are connected to the physical link 1. Although the connections from the VLANs are shown as wires 20, it will be understood that all the members of the multiplicity of VLANs are physically coupled to the link 1 but that it is possible to distinguish, as indicated above, between traffic from the various VLANs. Thus although the counters 11, 12 and 13 are shown as coupled to separate ones of the wires 20, in reality there is a single physical connection coupled to a filtering function to distinguish between traffic from the various VLANs and respective counters which count the filtered traffic counts.

[0024] The purpose of the monitoring function is to generate a priority decision signal (P_i). If ‘priority’ is true, then the associated VLAN is below its quota and the priority decision signal or priority flag is asserted.

[0025] Each VLAN has a request signal R_iwhich is asserted when that VLAN wishes to send packet over the link 1. The set of signals R_ifrom the VLANs are called the ‘restricted’ set. These are received as inputs to a selector 22. The ‘restricted set’ are also respectively gated with the priority flags from the monitors 11, 12 and 13 so that the output from the respective one of the AND gates 23 is asserted if a request signal is asserted and the respective monitor has asserted a priority flag. The outputs from the AND gates constitute a second set of inputs, the set of ‘priority requests’, to the selector.

[0026] The AND gate outputs are monitored, for example by an OR gate 24, so as to determine whether any of the priority requests are asserted. The selector is arranged so that if there is an output from the OR gate, indicating that at least one of the VLANs is in a priority state and is requesting access, the selector selects the priority signals, that is to say it arbitrates only between those VLANs which are in the priority state. If however no priority request is asserted, then the restricted set is chosen and the arbitration scheme arbitrates between the members of the restricted set. Thus, if there are no priority requests then the restricted sets will be searched. This means that the entire information carrying capacity can be utilised but that each VLAN is guaranteed its allocated quota.

[0027] If a VLAN is blocked from sending frames because it has exceeded its quota and there is not available spare bandwidth then packets may be lost. A VLAN may delay sending requests until its monitoring indicates that the VLAN is below its quota. Buffering may be employed, in conjunction with flow control if desired or necessary, to avoid excessive consumption of buffering capacity.