Multiple protocol engine transaction processing
A multi-processor computer system is described in which transaction processing is distributed among multiple protocol engines. The system includes a plurality of local nodes and an interconnection controller interconnected by a local point-to-point architecture. The interconnection controller comprises a plurality of protocol engines for processing transactions. Transactions are distributed among the protocol engines using destination information associated with the transactions.
1. A method for processing memory transactions in a cluster having a plurality of processing nodes and an interconnection controller coupled to the plurality of processing nodes, the interconnection controller including a plurality of protocol engines, each engine configured to process multiple memory transactions wherein each of said plurality of protocol engines is configured to be assigned a disjoint subset of a global memory space, the method comprising:
receiving a first memory transaction with the interconnection controller;
identifying one of the disjoint subsets using destination information associated with the first memory transaction, the one of the identified subsets corresponding to one of remote and local memory;
selecting a first protocol engine of the plurality corresponding to the identified disjoint subset;
processing the memory transaction with the first selected protocol engine in accordance with a cache coherence protocol;
receiving a second memory transaction with the interconnection controller;
identifying another of the disjoint subsets using destination information associated with the second memory transaction, the other of the identified subsets corresponding to one of remote and local memory;
selecting a second protocol engine of the plurality corresponding to the other of the identified disjoint subsets; and
processing the second memory transaction with the second selected protocol engine in accordance with the cache coherence protocol.
2. The method of claim 1 , wherein destination information for each memory transaction comprises a target address contained in a packet associated with the memory transaction.
3. The method of claim 2 , wherein each selected protocol engine is selected with reference to at least one bit in the target address, said at least one bit corresponding to one of a memory line boundary and a memory page boundary.
4. The method of claim 1 , wherein the memory transactions have corresponding target addresses associated therewith, the method further comprising distributing the memory transactions for processing among the protocol engines by interleaving the target addresses.
5. A computer system, comprising:
(a) a plurality of processing nodes;
(b) an interconnection controller coupled to the plurality of processing nodes, the interconnection controller comprising a plurality of protocol engines, each configured to process multiple memory transactions therein in accordance with a cache coherence protocol, wherein each protocol engine of said plurality of protocol engines is configured to be assigned a substantially mutually-exclusive subset of a global memory space,
(c) means for distributing a first memory transaction& for processing by a first protocol engine of the plurality of protocol engines by identifying a subset of the global memory space using destination information associated with the first memory transaction, the subset corresponding to a remote memory or a local memory; and
(d) means for distributing a second memory transaction for processing by a second protocol engine of the plurality of protocol engines by identifying another subset of the global memory space using destination information associated with the second memory transaction, the other subset corresponding to a remote memory or a local memory.
6. The computer system of claim 5 wherein the interconnection controller further comprises a protocol engine configured to process interrupts, I/O accesses, and configuration accesses.
7. The computer system of claim 5 wherein the processing nodes and interconnection controller are interconnected in a point-to-point architecture having a topology selected from the group consisting of a ring topology, a mesh topology, and a switch topology.