Progressive adaptive routing in a dragonfly processor interconnect network

Abstract

A multiprocessor computer system comprises a dragonfly processor interconnect network that comprises a plurality of processor nodes and a plurality of routers. The routers are operable to adaptively route data by selecting from among a plurality of network paths from a target node to a destination node in the dragonfly network based on one or more of network congestion information from neighboring routers and failed network link information from neighboring routers.

What is claimed is: 
     
         1 . An apparatus comprising:
 at least one first interface to couple to one or more processors in a plurality of processors;   at least one second interface to couple to a plurality of routers in a particular group of routers, wherein the particular group is one of a plurality of groups in a dragonfly topology, each of the groups in the plurality of groups comprises a respective plurality of routers, and each router is to be connected to every other router in its group;   at least one third interface to support an optical channel to couple the particular group to one other group in the plurality of groups, wherein each group in the plurality of groups is connected to each other group in the plurality of groups by an optical channel; and   routing logic to use at least one of a plurality of routing tables for use in routing data between processors in the plurality of processors.   
     
     
         2 . The apparatus of  claim 1 , wherein the first interface comprises one or more ports, the second interface comprises one or more ports, and the third interface comprises one or more optical ports. 
     
     
         3 . The apparatus of  claim 1 , wherein data is to be routed over at most two local channel hops and at most one global optical channel hops. 
     
     
         4 . The apparatus of  claim 1 , wherein at least one of the routing tables facilitates minimal routing. 
     
     
         5 . The apparatus of  claim 1 , wherein at least one of the routing tables facilitates non-minimal routing. 
     
     
         6 . The apparatus of  claim 1 , wherein at least one of the routing tables facilitates adaptive routing. 
     
     
         7 . The apparatus of  claim 6 , wherein adaptive routing balances load across global channels that interconnect groups in the plurality of groups. 
     
     
         8 . The apparatus of  claim 1 , wherein at least one of the routing tables is used for routing between groups and at least one of the routing tables is used for routing between router modules within a particular one of the plurality of groups. 
     
     
         9 . A system comprising:
 a plurality of processors nodes; and   a plurality of router modules,   wherein each of the plurality of router modules are coupled to one or more of the plurality of processor nodes, the plurality of router modules comprise a dragonfly topology network comprising a plurality of groups, each group in the plurality of groups comprises a respective subset of the plurality of router modules, the router modules of each group are interconnected to each other router module in the group by a respective local channel, each group in the plurality of groups is interconnected with each other group in the plurality of groups by a respective optical channel, and one or more routing tables are to be associated with each of the plurality of groups.   
     
     
         10 . The system of  claim 9 , further comprising routing logic to use the routing tables to route data between two of the plurality of router modules. 
     
     
         11 . The system of  claim 9 , wherein the plurality of router modules form one or more high radix routers. 
     
     
         12 . The system of  claim 9 , wherein data is to be routed over at most two local channel hops and at most one global optical channel hops. 
     
     
         13 . The system of  claim 9 , wherein at least one of the routing tables facilitates minimal routing. 
     
     
         14 . The system of  claim 9 , wherein at least one of the routing tables facilitates non-minimal routing. 
     
     
         15 . The system of  claim 9 , wherein at least one of the routing tables facilitates adaptive routing. 
     
     
         16 . The system of  claim 9 , wherein adaptive routing balances load across global channels that interconnect groups in the plurality of groups. 
     
     
         17 . The system of  claim 9 , wherein at least one of the plurality of groups corresponds to a server chassis. 
     
     
         18 . The system of  claim 17 , further comprising a plurality of server chassis, wherein global optical channels interconnect the server chassis with at least one other server chassis. 
     
     
         19 . The system of  claim 9 , wherein at least one of the routing tables is used for routing between groups and at least one of the routing tables is used for routing between router modules within a particular one of the plurality of groups. 
     
     
         20 . An apparatus comprising:
 a plurality of processors;   a plurality of router modules to interconnect in a dragonfly topology, wherein the dragonfly topology is to comprise a plurality of groups, each router module comprises:
 at least one first interface to couple the router module to one or more of the plurality of processors; 
 at least one second interface to couple the router module to every other router module in its respective group of router modules; and 
 at least one third interface to couple the group of router modules of the router module to another one of the plurality of groups of router modules over a global optical channel, and 
 routing logic to use one or more of a plurality of routing tables to route data. 
   
     
     
         21 . The apparatus of  claim 20 , wherein data is to be routed over at most two local channel hops and at most one global optical channel hops. 
     
     
         22 . The apparatus of  claim 20 , wherein the first interface comprises one or more ports, the second interface comprises one or more ports, and the third interface comprises one or more optical ports. 
     
     
         23 . The apparatus of  claim 20 , wherein at least one of the routing tables facilitates minimal routing. 
     
     
         24 . The apparatus of  claim 20 , wherein at least one of the routing tables facilitates non-minimal routing. 
     
     
         25 . The apparatus of  claim 20 , wherein at least one of the routing tables facilitates adaptive routing. 
     
     
         26 . The apparatus of  claim 20 , wherein at least one of the routing tables is used for routing between groups and at least one of the routing tables is used for routing between router modules within a particular one of the plurality of groups.