Network topology for a scalable multiprocessor system
A system and method for interconnecting a plurality of processing element nodes within a scalable multiprocessor system is provided. Each processing element node includes at least one processor and memory. A scalable interconnect network includes physical communication links interconnecting the processing element nodes in a cluster. A first set of routers in the scalable interconnect network route messages between the plurality of processing element nodes. One or more metarouters in the scalable interconnect network route messages between the first set of routers so that each one of the routers in a first cluster is connected to all other clusters through one or more metarouters.
1. A parallel processing system, comprising:
a plurality of clusters, including at least a first cluster, a second cluster, and a third cluster, wherein each cluster of the plurality of clusters includes:
a plurality of processing element nodes, each of the plurality of nodes including at least one processor, and
a first set of routers for interconnecting the plurality of processing element nodes in each cluster of the plurality of clusters; and
a second set of routers for interconnecting each of the clusters of the plurality of clusters, wherein:
a first message from a node in the first cluster is routed to the third cluster through a router of the second set of routers without being routed to the second cluster when the first message addresses a node in the third cluster,
a second message from a node of the plurality of nodes in the second cluster is routed to the third cluster through the router or through another router of the second set of routers without being routed through the first cluster when the second message addresses the node or another node in the third cluster.
2. The parallel processing system of claim 1 , wherein a router in each of the plurality of clusters has a direct connection to a router in at least one other cluster in the plurality of clusters.
3. The parallel processing system of claim 1 , wherein a router of the first set of routers in each of the plurality of clusters is connected via a router in the second set of routers to a router in at least one other cluster of the plurality of clusters.
4. The parallel processing system of claim 1 , wherein each of the clusters is a two-dimensional hypercube.
5. The parallel processing system of claim 1 , wherein each of the clusters is a three-dimensional hypercube.
6. The parallel processing system of claim 1 , wherein each router in the second set of routers is an eight port router.
7. The parallel processing system of claim 1 , wherein each router in the second set of routers is a four port router.
8. The parallel processing system of claim 1 , wherein the plurality of processing element nodes include four processors.
9. The parallel processing system of claim 1 , wherein the plurality of clusters and second set of routers form a scalable network, wherein the clusters include n-dimensional hypercube that remain intact as additional processing elements nodes are added to the scalable network.
10. A method for providing parallel processing in a plurality of clusters including at least a first cluster, a second cluster, and a third cluster, wherein each cluster of the plurality of clusters includes a plurality of processing element nodes and a first set of routers, the method comprising:
interconnecting a plurality of nodes in each cluster of the plurality of clusters by way of the first set of routers;
interconnecting each of the clusters of the plurality of clusters by way of the second set of routers; and
connecting each of the plurality of clusters, such that each of the plurality of nodes in each respective cluster of the plurality of clusters are connected through a router of the first set of routers in each respective cluster, and each of the clusters of the plurality of clusters are connected by at least one router of the second set of routers,
routing a first message from a node in the first cluster to the third cluster through a router of the second set of routers without being routed to the second cluster when the first message addresses a node in the third cluster,
routing a second message from a node of the plurality of nodes in the second cluster to the third cluster through the router or through another router of the second set of routers without being routed through the first cluster when the second message addresses the node or another node in the third cluster.
11. The method of claim 10 , further comprising establishing a direct connection by a router in each of the plurality of clusters to a router in at least one other cluster of the plurality of clusters.
12. The method of claim 10 , wherein at least one router of the first set of routers in each of the plurality of clusters is connected to a router in each of the other clusters of the plurality of clusters via a router in the second set of routers.
13. The method of claim 10 , further comprising wherein each of the clusters is a two dimensional hypercube.
14. The method of claim 10 , further comprising wherein each of the clusters is a three dimensional hypercube.
15. The method of claim 10 , further comprising adding additional processing elements nodes to a scalable network, wherein the plurality of clusters and second set of routers form the scalable network, wherein the clusters include n-dimensional hypercube that remain intact as the additional processing elements nodes are added to the scalable network.
16. The method of claim 10 , wherein each router in the second set of routers is an eight port router.
17. The method of claim 10 , wherein each router in the second set of routers is a four port router.
18. The method of claim 10 , wherein the plurality of processing element nodes include four processors.