Method and apparatus for preventing loops in a network by controlling broadcasts

This application is a continuation of U.S. application Ser. No. 11/278,314, filed Mar. 31, 2006, the entirety of which is incorporated herein by reference.

1. Field of Invention

The present invention relates generally to communication networks. More particularly, the present invention relates to methods of fault protection in communication networks.

2. Description of the Related Art

Various network topologies utilize rings. Resilient Packet Ring (RPR) is an example of a network topology for fiber optic rings. The Institute of Electrical and Electronic Engineers (IEEE) defines RPR standards in IEEE 802.17, which is incorporated herein by reference in its entirety. RPR is a layer two ring technology in a networking framework defined by the Open Systems Interconnection (OSI) model, and supports bi-directional communications around both sides of a ring using redundant paths substantially simultaneously. At layer two, or a data link layer, data packets are encoded and decoded into bits. Layer two is divided into two sublayers, a Media Access Control (MAC) layer and a Logical Link Control (LLC) layer. The MAC sublayer is arranged to substantially control how a node of a network accesses and transmits data, while the LLC layer generally controls frame synchronization, controls flow control, and provides error checking.

RPR generally treats a ring as a broadcast medium, and does not specify a higher level loop prevention mechanism. When two rings are in communication over interconnects, looping may occur. FIG. 1A is a diagrammatic representation of two RPR clusters that are in communication over a set of interconnects. A first RPR cluster 110 a and a second RPR cluster 110 b are in communication over interconnects 114 a - d . More than one interconnect 114 a - d is typically present between first RPR cluster 110 a and second RPR cluster 110 b to avoid a single point of failure.

When first RPR cluster 110 a and second RPR cluster 110 b choose the same interconnect, i.e., interconnect 114 c as shown in FIG. 1B , to use for communications, then looping is not likely to occur. However, when first RPR cluster 110 a and second RPR cluster 110 b select different interconnects 114 a - d to communicate over, looping may occur. FIG. 1C is a diagrammatic representation of first RPR cluster 110 a that sends messages to second RPR cluster 110 b over interconnect 114 d and second RPR cluster 110 b sending messages to first RPR cluster 110 a over interconnect 114 b . Looping may occur, as when first RPR cluster 110 a receives a message for broadcast on interconnect 114 b , first RPR cluster 110 a may broadcast the message back to second RPR cluster 110 b on interconnect 114 d . Such looping may flood interconnects 114 b , 114 d , and degrade the performance of interconnects 114 b , 114 d , as well as of first RPR cluster 110 a and second RPR cluster 110 b . Looping may also result in the replication of frames contained in messages when a frame broadcast on interconnect 114 b is returned on interconnect 114 d.

Looping will be described in more detail with respect to FIG. 2 . Multiple active paths between rings often cause loops in between the rings. When a loop is in existence in a network, duplication of messages may occur, as some nodes effectively receive messages from different sides. By way of example, when a node is a bridge, looping results in the bridge receiving a message on both sides of the bridge. As a result, a message forwarding algorithm of the bridge may become confused, and duplicate frames may be forwarded.

FIG. 2 is a diagrammatic representation of a network which includes nodes arranged in two rings, i.e., a bridged RPR network with a dual ring interconnect. A first ring includes nodes 206 a - h , and a second ring includes nodes 216 a - h . When a customer 202 wishes to send a message on layer two, as for example a message that is arranged as a plurality of frames, to another customer 220 , the message is propagated through the first ring and the second ring. As the rings are linked by interconnects 210 , 212 , the message may loop through nodes 206 a - h . Once the message reaches nodes 206 c , 206 g , the message is sent on interconnects 212 , 210 , respectively, to nodes 216 c , 216 g , respectively. Both node 216 c and node 216 g propagate the message through the second ring and, eventually, to customer 220 through node 216 e.

However, when the version of the message received on node 216 c via interconnect 212 is forwarded to node 216 g , node 216 g may forward that version of the message through interconnect 210 back to node 206 g . Similarly, when the version of the message received on node 216 g via interconnect 210 is forwarded to node 216 c , node 216 c may forward that version of the message through interconnect 212 back to node 206 c . Hence, the message is effectively looped between the first ring and the second ring.

A spanning tree protocol may be used in some instances to effectively guarantee a loop-free network. To avoid path redundancy, a spanning tree protocol defines a tree that spans all nodes in a network. A spanning tree protocol effectively causes some redundant interconnects to be blocked. If one unblocked interconnect fails or otherwise becomes unavailable, a spanning tree algorithm reconfigures the network topology by unblocking a blocked path. However, a spanning tree protocol is relatively slow in recovering from a fault. In other words, once a fault is discovered, although a spanning tree protocol may be effective in reconfiguring a system, the reconfiguration process is relatively slow. The performance of the system may be compromised by a slow recover reconfiguration or recover process.

Though a rapid spanning tree protocol, which is implemented in some networks to guarantee a loop-free network, generally provides a faster reconfiguration process than a spanning tree protocol, a rapid spanning tree protocol does not provide rapid reconfiguration in a ring. Hence, a rapid spanning tree protocol may not be used to ensure that connectivity between rings is restored after a fault occurs.

Therefore, what is needed is a method and an apparatus for allowing a network that includes rings to operate without loops and to be relatively rapidly reconfigured in the event of a failure of an interconnect between the rings. That is, what is desired is a system that prevents loops from forming between rings, and that ensures a relatively efficient restoration of connectivity between the rings in the event of a failure.

The invention may best be understood by reference to the following description taken in conjunction with the accompanying drawings in which:

FIG. 1A is a diagrammatic representation of a network in which a first resilient packet ring (RPR) cluster is interconnected with a second RPR cluster using multiple interconnecting links.

FIG. 1B is a diagrammatic representation of a first RPR cluster and a second RPR cluster, i.e., first RPR cluster 110 a and second RPR cluster 110 b of FIG. 1A , communicating bi-directionally over a single link.

FIG. 1C is a diagrammatic representation of a first RPR cluster and a second RPR cluster, i.e., first RPR cluster 110 a and second RPR cluster 110 b of FIG. IA, communicating over a plurality of links.

FIG. 2 is a diagrammatic representation of a plurality of clusters in between which loops may occur.

FIG. 3A is a diagrammatic representation of a parent cluster and a child cluster which are in communication over multiple interconnecting links in accordance with an embodiment of the present invention.

FIG. 3B is a diagrammatic representation of a parent cluster, i.e., parent cluster 310 a of FIG. 3A , in communication with a plurality of subclusters of a child cluster, i.e., child cluster 310 b of FIG. 3A , over multiple interconnecting links in accordance with an embodiment of the present invention.

FIG. 3C is a diagrammatic representation of a parent cluster, i.e., parent cluster 310 a of FIGS. 3A and 3B , broadcasting a message to a plurality of subclusters of a child cluster, i.e., child cluster 310 b of FIGS. 3A and 3B , over multiple interconnecting links in accordance with an embodiment of the present invention.

FIG. 3D is a diagrammatic representation of each subcluster of a child cluster, i.e., child cluster 310 b of FIG. 3C , broadcasting separate messages to a parent cluster, i.e., parent cluster 310 a of FIG. 3C , over specific interconnecting links in accordance with an embodiment of the present invention.

FIG. 4A is a diagrammatic representation of a parent cluster with a filter and a child cluster with a filter in accordance with an embodiment of the present invention.

FIG. 4B is a diagrammatic representation of a node with a filter and frame broadcasting mechanism in accordance with an embodiment of the present invention.

FIG. 5 is a process flow diagram which illustrates one method of configuring a network to have a designated parent cluster and a designated child cluster in accordance with an embodiment of the present invention.

FIG. 6 is a process flow diagram which illustrates one method of processing a message sent from a parent cluster to a child cluster via links associated with subclusters in accordance with an embodiment of the present invention.

FIG. 7 is a process flow diagram which illustrates one method of sending a message from a child cluster to a parent cluster via a link associated with a subcluster in accordance with an embodiment of the present invention.

FIG. 8A is a diagrammatic representation of network in which a child cluster is divided into subclusters based on the physical location of nodes in accordance with an embodiment of the present invention.

FIG. 8B is a diagrammatic representation of a network in which a child cluster is divided into subclusters based on traffic distribution in accordance with an embodiment of the present invention.

FIG. 9 is a diagrammatic representation of a network in which traffic is routed using spatial balancing in accordance with an embodiment of the present invention.

FIG. 10 is a diagrammatic representation of a network in which traffic is routed using virtual local area network (VLAN) balancing in accordance with an embodiment of the present invention.

To provide fast recovery of network faults, multiple communication paths are provided for every possible communication to effectively ensure that the failure of a single underlying communication path or device does not cause the failure of the communication to propagate through the entire network. Establishing multiple communication paths, or interconnects, between areas of a network, e.g., clusters, creates communication loops. Communication loops may cause communication to propagate multiple times through the same devices or communication paths, resulting in the duplication of communication messages and crippling use of bandwidth.

In one embodiment, a network that is partitioned into a plurality of clusters, as for example rings of a Resilient Packet Ring (RPR) network, uses substantially all interconnects for communication without allowing communication loops, while supporting a relatively rapid recovery of all communication paths in the event of failure of an interconnect. A parent cluster is defined to broadcast messages on substantially all available interconnects between the parent cluster and an associated child cluster. The child cluster is defined to have multiple subclusters, each of which is associated with an interconnect. Looping may be prevented, as the child cluster broadcasts messages to the parent cluster on only one interconnect at any given time. Hence, configuring the parent cluster to identify that a message is received on a given interconnect may prevent the subcluster that originated the message from having the message looped back to it by the parent cluster. Typically, topology information does not need to be communicated between the parent cluster and the child cluster.

Loops may be prevented between a pair of clusters by sending communications on one or more interconnects between the clusters and essentially ensuring that the same communication does not return back on any of the interconnects. If a communication is sent over more than one interconnect, duplication of the communication in the destination cluster may be avoided by limiting the scoping of allowed destinations for each interconnect in the destination cluster.

Referring next to FIG. 3A , a parent cluster and a child cluster will be described in accordance with an embodiment of the present invention. Within a network 300 , e.g., a cluster interconnect group, clusters 310 a , 310 b , which may be rings, are in communication across multiple interconnects 314 a - d . In general, clusters 310 a , 310 b each include of multiple nodes which are able communicate bi-directionally within clusters 310 a , 310 b . By way of example, each node or device in a cluster 310 a is arranged to communicate bi-directionally with every other node or device in cluster 310 a . The bi-directional communication may be achieved through any suitable broadcast connection. Suitable broadcast connections include, but are not limited to, shared physical media, e.g., a shared Ethernet segment or wireless broadcast area, and a logical broadcast domain created using encapsulation such as an Institute of Electrical and Electronic Engineers (IEEE) 802.17 compliant ring or a multi-point to multi-point tunnel. Bi-directional communication within clusters 310 a , 310 b may also utilize connection forwarding, replication, and filtering that allows for the creation of logical broadcast domains overlaid on a form of a connection network. It should be appreciated that in addition to broadcast communication, unicast communication may also occur inside clusters 310 a , 310 b.

Interconnects 314 a - d are communication paths between devices in clusters 310 a , 310 b , and are typically point-to-point connections that support bi-directional communication. In other words, each interconnect 314 a - d is arranged between a device in cluster 310 a and a device in cluster 310 b , and may carry both broadcast and unicast communication. Generally, each interconnect 314 a - d is arranged between a different pair of devices. Interconnects 314 a - d may be physical links such as Ethernet links. Alternatively, interconnects 314 a - d may be logical communication paths over media, or essentially conceptual communication paths that are internal to a device which is a member of both clusters 310 a , 310 b.

In the described embodiment, cluster 310 a is a parent cluster and cluster 310 b is a child cluster 310 b . That is, the hierarchical relationship between clusters 310 a , 310 b is such that cluster 310 a is a parent and cluster 310 b is a child. Within parent cluster 310 a , nodes are configured, as for example by a network administrator, to support interconnects 314 a - d as being associated with a parent. Within child cluster 310 b , nodes are configured to support interconnects 314 a - d as being associated with a child. Further, child cluster 310 b is divided into subclusters, or cluster subsets, that are each associated with an interconnect 314 a - d , as will be described with respect to FIG. 3B .

FIG. 3B is a diagrammatic representation of parent cluster 310 a interconnected with child cluster 310 b which is divided into subclusters in accordance with an embodiment of the present invention. Child cluster 310 b is divided or partitioned into subclusters 318 a - d . Although approximately four subclusters 318 a - d are shown, it should be appreciated that another number of subclusters 318 a - d may generally be included in child cluster 310 b . A subcluster 318 a - d may include all nodes 322 of child cluster 310 b , no nodes 322 , or some nodes 322 . Subclusters 318 a - d are defined such that all nodes 322 are encompassed, i.e., all nodes 322 are accounted for as a part of a subcluster 318 a - d . Nodes 322 are such that any given node 322 is included in only one subcluster 318 a - d . It should be appreciated that a subcluster 318 a - d may include no nodes if an associated interconnect 314 a - d is faulty or otherwise offline.

Each interconnect 314 a - d is associated with a subcluster 318 a - d . As shown, interconnect 314 a is associated with subcluster 318 a , interconnect 314 b is associated with subcluster 318 b , interconnect 314 c is associated with subcluster 318 c , and interconnect 314 d is associated with subcluster 318 d . Generally, subclusters 318 a - d may be dynamically allocated to interconnects 314 a - d such that a high level of performance may be achieved. The allocation of nodes 322 within subclusters 318 a - d , or the determination of subclusters 318 a - d , will be described below with respect to FIGS. 8A and 8B .

When parent cluster 310 a broadcasts a message to child cluster 310 b , the message is broadcast on all interconnects 314 a - d . As shown in FIG. 3C , a message 330 that is broadcast by parent cluster 310 a is sent on all interconnects 314 a - d to child cluster 310 b . Although not shown, it should be understood that when parent cluster 310 a unicasts a message to a particular node 322 in child cluster 310 b , the unicast message may be sent on the appropriate interconnect 314 a - d to the subcluster 318 a - d in which node 322 is resident.

Interconnects 314 a - d generally have the ability to support a subcluster broadcast to all nodes 322 in an associated subcluster 318 a - d , respectively. While each interconnect 314 a - d is configured to support a single subcluster 318 a - d , respectively, at any given time, it should be understood that the subcluster 318 a - d to which an interconnect 314 a - d may change, e.g., when one interconnect 314 a - d fails and thereby causes subclusters 318 a - d to be redefined. That is, each interconnect 314 a - d may be arranged to be associated with any subcluster 318 a - d . Each interconnect 314 a - d may, in one embodiment, support the ability to broadcast to substantially only nodes 322 included in its associated subcluster 318 a - d , respectively, for some communications and to broadcast other communications to all nodes 322 in child cluster 310 b.

To prevent looping from occurring, when subclusters 318 a - d or, more specifically, nodes 322 within subclusters 318 a - d communicate messages to parent cluster 310 a , e.g., through unicast or broadcast, the messages are communicated substantially only on the interconnect 314 a - d which is associated with the subcluster 318 a - d originating the communication within child cluster 310 b . By way of example, when subcluster 318 a broadcasts a message, the message is broadcast to substantially all subclusters 318 - d in child cluster 310 b , but is allowed to reach parent cluster 318 a substantially only by way of interconnect 314 a and not on other interconnects 314 b - d . Looping is prevented by parent cluster 310 a or, more specifically, each node which controls an interconnect 314 a - d , being aware that the message was received by way of an interconnect 314 a - d , and preventing the message from returning on any interconnect 314 a - d.

FIG. 3D is a diagrammatic representation of each subcluster 318 a - d broadcasting messages to parent cluster 310 a associated interconnects 314 a - d , respectively, in accordance with an embodiment of the present invention. When subcluster 318 a or, more specifically, a node 322 associated with subcluster 318 a , broadcasts a message 340 a to parent cluster 310 a , message 340 a is substantially only sent on interconnect 314 a . Message 340 a is not sent to parent cluster 310 a over interconnects 314 b - d . When a node in subcluster 318 b broadcasts a message 340 b to parent cluster 310 a , message 340 b is broadcast substantially only on interconnect 314 b . Similarly, a message from subcluster 318 c is broadcast to parent cluster 310 a substantially only on interconnect 314 c , and a message from subcluster 318 d is broadcast to parent cluster 310 a substantially only on interconnect 314 d.

To control communication between clusters 310 a , 310 b , and to enable subclusters 318 a - d to be associated with particular interconnects 314 a - d , filters may be implemented. Each interconnect 314 a - d supports the ability to filter communications to and from the child cluster 310 b . FIG. 4A is a diagrammatic representation of a parent cluster with a filter and a child cluster with a filter in accordance with an embodiment of the present invention. A parent cluster 410 a is in communication with a child cluster 410 b over interconnects 414 a , 414 b . Child cluster 410 b includes two subclusters 418 a , 418 b of nodes (not shown). Parent cluster 410 includes at least one node 422 which, in the described embodiment, is arranged to broadcast a message to child cluster 410 b.

A subcluster filter arrangement 480 is arranged to be bi-directional, and to filter both signals transmitted and received between interconnects 414 a , 414 b and child cluster 410 b . Typically, subcluster filter arrangement 480 has filter function that is distributed to nodes (not shown) that are attached to an interconnect. Subcluster filter arrangement 480 may be implemented as code devices, e.g., computer code that executes on a processor of a computing system, or as hardware, e.g., as a part of a line card. When subcluster filter arrangement 480 is in an enabled state, substantially all incoming communications from parent cluster 410 a that are not locally defined in a subclusters 418 a , 418 b are blocked from being received by subclusters 418 a , 418 b . By way of example, if node 422 sources a unicast message that is intended for subcluster 418 b , a filter arrangement 482 of parent cluster 410 a may send the unicast message on interconnect 414 b , and subcluster filter arrangement 480 may make certain that subcluster 418 b receives the unicast message while subcluster 418 a does not. Filter arrangement 482 may be arranged such that interconnects 414 a , 414 b may identify which subcluster 418 a , 418 b contains a node (not shown) that sourced a message to be sent to parent cluster 410 a . Identifying which subcluster 418 a , 418 b a message was sourced from prevents the message from being sent back to the subcluster 418 a , 418 b , i.e., loops may be prevented.

In one embodiment, a filter arrangement such as filter arrangement 480 or 482 may be included in a node, e.g., a node that is in a child cluster 410 b or a parent cluster 410 a . That is, filter functionality may be distributed to each node attached to an interconnect such that a single point of failure is effectively eliminated. As shown in FIG. 4B , a node 490 that is a part of a cluster 492 may include a filter 494 and a frame broadcasting mechanism 496 that is arranged to broadcast a frame on at least one interconnect 498 .

With reference to FIG. 5 , one method of configuring an interconnect group and communicating within the interconnect group will be described in accordance with an embodiment of the present invention. A process 500 of configuring an interconnect group of a network begins at step 504 in which a parent cluster and a child cluster are identified. In one embodiment, the parent cluster and the child cluster are resilient packet rings that are in communication over at least one interconnect. Once a parent cluster and a child cluster are identified, the parent cluster is configured to broadcast to the child cluster on substantially all interconnects of a pool of available, or working, interconnects in step 508 . That is, the parent cluster is arranged such that when the parent cluster broadcasts a message to the child cluster, the message is broadcast on substantially all available interconnects between the parent cluster and the child cluster. A network administrator or software may configure nodes or devices within the parent cluster to broadcast messages on substantially all available interconnects.

In step 512 , the child cluster is configured and divided into subclusters. Dividing the child cluster into subclusters may include, but is not limited to, grouping nodes in the child cluster according to geographic location, or identifying traffic demands and grouping nodes into subclusters accordingly. After the child cluster is divided into subclusters, each subcluster is configured in step 516 to receive messages and to broadcast messages on a designated interconnect. In one embodiment, each subcluster is assigned an interconnect that the subcluster uses to receive and to broadcast messages. In general, each interconnect of a pool of available interconnects is assigned to a single subcluster.

From step 516 , process flow moves to step 520 in which messages, e.g., broadcast or unicast, are sent and received on the interconnects. A determination is made in step 524 as to whether any of the interconnects has failed or is otherwise unavailable for use in transporting messages between the parent cluster and the child cluster. If it is determined that no interconnects have failed, process flow returns to step 520 in which messages are transported across the interconnects. Alternatively, if it is determined in step 524 that an interconnect has failed, then in step 528 , the failed interconnect is removed from the pool of available interconnects. Once the failed interconnect is removed from the pool of available interconnects, process flow returns to step 508 in which the parent cluster is effectively reconfigured to broadcast on all remaining interconnects in the pool of available interconnects. In one embodiment, reconfiguring a system after a failed interconnect has been detected may take less than approximately fifty milliseconds (ms).

Broadcasting messages from a parent cluster to a child cluster, as mentioned above, involves broadcasting messages on substantially all available interconnects or links between the parent cluster and the child cluster. FIG. 6 is a process flow diagram which illustrates one method of processing a message sent from a parent cluster to a child cluster via interconnects effectively assigned to subclusters in accordance with an embodiment of the present invention. A process 600 of processing a message begins at step 604 in which a message is broadcast from a parent cluster to a child cluster across interconnects, e.g., link X and link Y. In the described embodiment, a pool of available interconnects between the parent cluster and the child cluster includes link X and link Y.

Subcluster X, which is part of the child cluster, receives the broadcasted message on link X in step 608 . It should be appreciated that link X is the interconnect assigned to subcluster X. Similarly, subcluster Y, which is also part of the child cluster, receives the broadcasted message on link Y in step 612 . Subcluster X and subcluster Y may receive the broadcasted message at approximately the same time, or at slightly different times in the event that one subcluster is at a greater distance from the parent cluster than the other subcluster.

Upon receiving the broadcasted message, the receiving node or element in subcluster X propagates the message through subcluster X in step 620 . In other words, the node which receives the broadcasted message on link X effectively broadcasts the message to other nodes, if any, within subcluster X. In step 620 , the receiving node or element in subcluster Y propagates the message through subcluster Y. Generally, in spatial balancing, a frame may flow through substantially only those nodes that belong to a subcluster of a parent interconnect when the frame is being sent from the parent to a child. However, in logical balancing, a frame may flow through substantially all nodes in all subclusters, but may be ignored by nodes that do not belong to the subcluster on which the frame was received. Once the broadcasted message has been received and passed through subcluster X and subcluster Y, the process of processing a message is completed. It should be appreciated that the broadcasted message may, in some instances, be broadcasted through other interconnects to other clusters until all clusters that are intended to receive the message actually receive the message.

When a subcluster broadcasts a message to a parent cluster, the message is broadcast on the interconnect between the subcluster and the parent cluster. Referring next to FIG. 7 , one method of processing a message that is broadcast from a subcluster to a parent cluster will be described in accordance with an embodiment of the present invention. A process 700 of processing a message broadcast from a subcluster to a parent cluster begins at step 704 in which a message is broadcast from subcluster X across an associated interconnect, e.g., link X, to the parent cluster. In the described embodiment, the message is substantially only broadcast on link X. After the broadcast message is sent on link X, the parent cluster receives the message on link X in step 708 . Once the message is received, the parent cluster broadcasts the received message in step 712 throughout the parent cluster, but not through any interconnects. In general, the parent cluster may determine that if any matched nodes of the parent cluster received the message from a child cluster, the message is not to be forwarded back to the child cluster from which the message was received. That is, a frame is broadcast such that it does not return to the child cluster that sent the frame. Upon broadcasting the receive message throughout the parent cluster, the process of processing a message that is broadcast from a subcluster is completed.

In general, subclusters may be determined using any suitable method. Suitable methods, as previously mentioned, include, but are not limited to, allocating nodes to subclusters based on physical location and allocating nodes to subclusters based on traffic to the nodes. With reference to FIG. 8A , the allocation of nodes to subclusters based on physical location will be described in accordance with an embodiment of the present invention. An interconnect system 800 includes a parent cluster 810 a that is in communication with a child cluster 810 b over interconnects 814 a - c . Each subcluster 818 a - c may be defined to include nodes 822 that are in physical proximity to each other and to an appropriate interconnect 814 a - c . For example, nodes 822 a , 822 b may be allocated to subcluster 818 a if nodes 822 a , 822 b are in proximity to each other and to interconnect 814 a . Similarly, nodes 822 c , 822 d may be allocated to subcluster 818 c if nodes 822 c , 822 d are in proximity to each other and to interconnect 814 c.

Alternatively, in lieu of allocating nodes to subclusters based on physical location, nodes may be allocated to subclusters such that the traffic on interconnects is relatively evenly divided. That is, nodes may be assigned to subclusters such that each subcluster in a child cluster is associated with approximately the same amount of traffic. FIG. 8B is a diagrammatic representation of a network in which a child cluster is divided into subclusters based on traffic distribution in accordance with an embodiment of the present invention. Within an interconnect system 850 , a parent cluster 860 a is interconnected with a child cluster 860 b across interconnects 864 a - c . Traffic associated with nodes 825 , 827 , 829 may be such that the majority of traffic is associated with a single node 827 . Hence, node 827 may be assigned as the only node in a subcluster 872 b and, hence, the only node 827 included cluster 860 b that utilizes interconnect 864 b . Other nodes 825 , 829 may be divided such that nodes 825 are included in a subcluster 872 a and nodes 829 are included in subcluster 872 c . In the described embodiment, traffic associated with all nodes 825 is approximately the same as the traffic associated with node 827 , and also approximately the same as traffic associated with all nodes 829 . Therefore, interconnects 864 a - c are each arranged to carry substantially the same traffic load.

It should be understood that subclusters 872 a - c may be redefined at substantially any time. As the assignment of nodes to subclusters 872 a - c is dynamic, when traffic distribution changes, subclusters 872 a - c may be redefined to once again relatively evenly distribute traffic among subclusters 872 a - c . Redefining subclusters 872 a - c typically involves reallocating nodes 825 , 827 , 829 to subclusters 872 a - c.

Generally, subclusters may also be redefined or reconfigured when an interconnect has failed, as mentioned above with respect to FIG. 5 . In one embodiment, a failed interconnect is associated with a subcluster that is empty. Hence, when an interconnect is identified as faulty or failed, any nodes in the subcluster associated with the interconnect are disassociated from the interconnect. Typically, to maintain good performance within an interconnect group, nodes originally in a subcluster for which an interconnect is identified as faulty are relatively evenly allocated among other subclusters in a child cluster. Alternatively, in lieu of reallocating the nodes of a subcluster with a failed interconnect to other subclusters, completely new subclusters may be defined.

In one embodiment, two methods may be used to balance traffic between a parent cluster and a child cluster while effectively preventing replication of messages transmitted therebetween. Balancing traffic generally involves load sharing with a goal, for example, to ensure use of substantially all bandwidth in a ring. Bandwidth may be efficiently allocated by balancing traffic such that traffic flows substantially only to those nodes within a community of interest, such that bandwidth is not wasted on nodes that are not within the community of interest. Two possible methods used to balance traffic over interconnects are a spatial balancing method and a logical balancing method. Spatial balancing generally involves two nodes forwarding traffic based on topology of a network. Logical balancing generally involves each interconnect for subsets of logically identified communication, for example, communication of specific virtual local area networks (VLANs). Logical balancing substantially requires the identification of subclusters based on logical communication attributes such as VLAN association, and may involve independent identification of subclusters for each logical communication type, as long as the identification effectively prevents a single message coming from a parent cluster from being received more than once by nodes in a child cluster.

FIG. 9 is a diagrammatic representation of a network in which traffic is routed using spatial balancing in accordance with an embodiment of the present invention. In a network 900 , e.g., a bridged network or an RPR network, a node 902 may send a frame that is destined for a node 920 . Node 902 may generally be a first customer site node, while node 920 may generally be a second customer site node.

A frame is sent from node 902 via a path segment 905 a to a node 906 a which forwards the frame on paths 905 b , 905 c . Once the frame reaches nodes 906 b , 906 c , the frame is forward on interconnects 910 , 912 to nodes 916 a , 916 b . In the described embodiment, nodes 906 b , 906 c may be included in a parent cluster while nodes 916 a , 916 b may be included in a child cluster. Node 916 a forwards the frame on paths 915 a , 915 d , while node 916 b forwards the frame on paths 915 b , 915 c . If nodes 916 a , 916 b are included in a child cluster, then node 916 a and nodes 916 c - e may be a part of one subcluster while node 916 b and nodes 916 f - h are a part of another subcluster. Node 916 c provides the frame on a path segment 915 e to node 920 .

At a first level, node 906 a forwards the frame received from node 902 , i.e., a single matched node forwards the frame. It should be appreciated that nodes 906 b , 906 c may be considered to be matched nodes. Nodes 916 a , 916 b may also be considered to be matched nodes. Flood scoping prevents replication of the frame, while bi-directional flooding allows the shortest path to be realized.

With respect to FIG. 10 , VLAN balancing will be described in accordance with an embodiment of the present invention. FIG. 10 is a diagrammatic representation of a network in which traffic is routed using virtual local area network (VLAN) balancing. Within a network 950 , a customer site node 952 may forward a frame 955 a to a node 956 a that propagates the frame to its associated nodes on pats 955 b , 955 c . When the frame reaches nodes 956 b , 956 c , the frame is provided via interconnects 960 , 962 to nodes 966 a , 966 b , respectively. Matched nodes 966 a , 966 b , which may be considered to be associated with a child cluster, filter the frame such that only one of matched node 966 a , 966 b forwards the frame. Each node 966 a , 966 b may be associated with a VLAN, and filters frames based on the VLANS. It should be appreciated that filtering may be associated with either a parent process or a child process, or both. However, in the described embodiment, filtering functionality is provided by a child cluster. As shown, load balancing is such that node 966 b forwards the frame while node 966 a does not. Once a node 966 c receives the frame, the frame is provided to customer site 970 .

Although only a few embodiments of the present invention have been described, it should be understood that the present invention may be embodied in many other specific forms without departing from the spirit or the scope of the present invention. By way of example, although nodes may generally be complaint with the IEEE 802.17 standard, nodes may not necessarily be associated with RPR. In other words, the present invention may be implemented with regards to clusters that are not RPR. Hence, though the present invention has generally been described as being suitable for use in forwarding with respect to layer two ring technology in a networking framework defined by the Open Systems Interconnection (OSI) model, the present invention may be generalized for forwarding in layer three as defined by the OSI model.

When links or nodes within a cluster fail, communication between nodes within the cluster may be lost. As will be appreciated by those skilled in the art, a cluster may use protection schemes to allow communications within the cluster to occur even in light of a failed link or node. Protection schemes may include, but are not limited to, schemes involving 1+1 protection, unidirectional path switched rings, and bidirectional path switched rings.

In general, nodes or devices in clusters may be any suitable network element. Suitable network elements include, but are not limited to, switches, database systems, and computing systems. As will be appreciated by those skilled in the art, computing systems may include processors, memories, input/output devices, and display devices.

The identification of clusters of nodes within a network and substantially all interconnects between clusters may be accomplished by a network administrator. Alternatively, the identification of clusters and interconnects may be accomplished by network-wide topology aware software without departing from the spirit or the scope of the present invention.

The present invention describes a method of preventing communication loops between a pair of clusters with multiple balanced interconnects. It should be understood that, in general, networks may be divided into more than two clusters, and that the present invention may be modified for use in such networks. Messages originating in, or entering, a cluster may propagate to substantially all interconnected clusters, and may continue to propagate, e.g., following interconnect and subcluster filtering rules. A message received from a child cluster may be propagated to substantially all interconnected clusters, but not back to the child cluster. A message received by a child subcluster may be propagated through substantially all interconnects which are associated with devices in the subcluster. Each subcluster may be responsible for propagating the message to interconnects associated with the subcluster. In general, a cluster may be a child of one cluster and a parent of another cluster. The methods of the present invention may be applied to substantially any network that has no cyclical communication paths which extend through three or more clusters.

While a parent cluster has generally been described as being associated with one child cluster, a parent cluster may be associated with more than one child cluster. As described above, each child cluster, is typically associated with a single parent cluster. It should be understood, however, that by providing functionality that would substantially prevent duplication of broadcasts, a child cluster may be associated with more than one parent cluster. In general, a cluster associated with more than one parent cluster provides a method to ensure that one and only one copy of each message is propagated to each parent. Because a message received from a parent is duplicated into each subcluster, a filter may be configured to propagate a single message from a child cluster to each parent cluster. Possible methods for performing such filtering may include, but are not limited to, interconnect filtering within a child cluster based on a source and destination parent cluster that would filter communication from one parent to another except via a single dynamically designated interconnect, or interconnect filtering that would enable communication from one parent to another over a plurality of balanced interconnects with a filtering method that effectively ensures that one and only one copy of each message is propagated to each parent.

The term “communication” as used above generally refers to bridged or routed packet-based communication. However, it should be appreciated that other forms of streaming non-packet communication may occur within and between clusters. Such other forms of streaming non-packet communication is typically bi-directional and arranged to support some form of broadcast to more than one device.

While interconnects have generally been described as separate entities, e.g., separate devices or separate Ethernet links, interconnects may be a part of a single entity. For example, a single device may effectively serve as multiple interconnects. In an embodiment in which one device is used to provide more than one interconnection between a parent cluster and a child cluster, each interconnection within the device may have an associated subcluster filter.

A message that travels from a parent cluster to a child cluster within an interconnect group is replicated by the parent cluster to each subcluster, and is broadcast substantially only within each subcluster. It should be appreciated, however, that a message that originates in a subcluster or enters a subcluster via a method other than an interconnect from a parent cluster is generally broadcast directly to the nodes in each of the subclusters of the child cluster, without passing through any interconnect associated with the interconnect group. Communication between devices or nodes in the same cluster generally occurs directly between the devices or the nodes, and not through another cluster.

The steps associated with the methods of the present invention may vary widely. Steps may be added, removed, altered, combined, and reordered without departing from the spirit of the scope of the present invention. Therefore, the present examples are to be considered as illustrative and not restrictive, and the invention is not to be limited to the details given herein, but may be modified within the scope of the appended claims.