Abstract: A multi-processor computer system is described in which transaction processing in each cluster of processors is distributed among multiple protocol engines. Each cluster includes a plurality of local nodes and an interconnection controller interconnected by a local point-to-point architecture. The interconnection controller in each cluster comprises a plurality of protocol engines for processing transactions. Transactions are distributed among the protocol engines using destination information associated with the transactions.

Abstract: Cache coherence directory eviction mechanisms are described for use in computer systems having a plurality of multiprocessor clusters. Interaction among the clusters is facilitated by a cache coherence controller in each cluster. A cache coherence directory is associated with each cache coherence controller identifying memory lines associated with the local cluster which are cached in remote clusters. A variety of techniques for managing eviction of entries in the cache coherence directory are provided.

Abstract: An interconnection controller for use in a computer system having a plurality of processor clusters is described. Each cluster includes a plurality of local nodes and an instance of the interconnection controller. The interconnection controller is operable to transmit locally generated interrupts to others of the clusters, and remotely generated interrupts to the local nodes. The interconnection controller is further operable to aggregate locally generated interrupt responses for transmission to a first remote cluster from which a first interrupt corresponding to the locally generated responses was generated. The interconnection controller is also operable to aggregate remotely generated responses for transmission to a first local node from which a second interrupt corresponding to the remotely generated responses was generated. A computer system employing such an interconnection controller is also described.

Abstract: A multi-processor computer system is described in which address mapping, routing, and transaction identification mechanisms are provided which enable the interconnection of a plurality of multi-processor clusters, wherein the number of processors interconnected exceeds limited address, node identification, and transaction tag spaces associated with each of the individual clusters.

Abstract: A multi-processor computer system is described in which address mapping, routing, and transaction identification mechanisms are provided which enable the interconnection of a plurality of multi-processor clusters, wherein the number of processors interconnected exceeds limited address, node identification, and transaction tag spaces associated with each of the individual clusters.

Abstract: A multi-processor computer system is described in which address mapping, routing, and transaction identification mechanisms are provided which enable the interconnection of a plurality of multi-processor clusters, wherein the number of processors interconnected exceeds limited address, node identification, and transaction tag spaces associated with each of the individual clusters.

Abstract: A performance monitor configured to count memory transactions and to issue an interrupt to the computer system if the monitor detects a specified number of transactions associated with a particular segment of the physical address space of the system. The monitor includes an interface suitable for coupling to an interconnect network of a computer system and configured to extract physical address information from a transaction traversing the interconnect network, a translation module adapted for associating the extracted physical address with one of a plurality of memory blocks and, in response thereto, incrementing a memory block counter corresponding to the memory block, and an interrupt unit configured to assert an interrupt if the block counter exceeds a predetermined value. The interface unit is configurable to selectively monitor either incoming or outgoing transactions and the translation unit preferably includes a plurality of region filters each comprising one or more of the memory blocks.

Abstract: A data processing system includes a plurality of processing nodes that each contain at least one processor and data storage. The plurality of processing nodes are coupled together by a system interconnect. The data processing system further includes a configuration utility residing in data storage within at least one of the plurality of processing nodes. The configuration utility selectively configures the plurality of processing nodes into either a single non-uniform memory access (NUMA) system or into multiple independent data processing systems through communication via the system interconnect.

Abstract: A performance monitor for a computer system that includes an interface, a filter module, and an address mapping module. The interface is suitable for coupling to an interconnect network of the computer system. The interconnect network links a local node of the system with at least one remote node of the system. The interface is configured to extract physical address information from a transaction traversing the interconnect network. The filter module associates the physical address with one of several memory blocks, where each memory block comprises a contiguous portion of the system's physical address space. The address mapping module associates the identified memory block with at least one range of virtual addresses associated with at least one of a plurality of concurrently executing programs and increments each of a set of access counters. The association between the selected memory block and the access counters is facilitated by a pointer field corresponding to the memory block.

Abstract: A method for avoiding livelocks due to colliding writebacks within a NUMA computer system is disclosed. The NUMA computer system includes at least two nodes coupled to an interconnect. Each of the two nodes includes a local system memory. In response to an attempt by a processor located at a home node to access a modified cache line at a remote node via a memory request at substantially the same time when a processor located at the remote node attempts to writeback the modified cache line to the home node, the writeback is allowed to complete at the home node without retry only if the writeback is from what a coherency directory within the home node considered as an owning node of the modified cache line. The memory request is then allowed to retry and completed at the home node.

Abstract: A non-uniform memory access (NUMA) computer system includes a plurality of processing nodes coupled to a node interconnect. The plurality of processing nodes include at least a remote processing node, which contains a processor having an associated cache hierarchy, and a home processing node. The home processing node includes a shared system memory containing a plurality of memory granules and a coherence directory that indicates possible coherence states of copies of memory granules among the plurality of memory granules that are stored within at least one processing node other than the home processing node. If the processor within the remote processing node has a reservation for a memory granule among the plurality of memory granules that is not resident within the associated cache hierarchy, the coherence directory associates the memory granule with a coherence state indicating that the reserved memory granule may possibly be held non-exclusively at the remote processing node.

Abstract: A method for avoiding livelocks due to colliding invalidating transactions within a non-uniform memory access system is disclosed. A NUMA computer system includes at least two nodes coupled to an interconnect. Each of the two nodes includes a local system memory. In response to a request by a processor of a first node to invalidate a remote copy of a cache line also stored within its cache memory at substantially the same time when a processor of a second node is also requesting to invalidate said cache line, one of the two requests is allowed to complete. The allowed request is the first request to complete without retry at the point of coherency, typically the home node. Subsequently, the other one of the two requests is permitted to complete.

Abstract: A method and system for providing an eviction protocol within a non-uniform memory access (NUMA) computer system are disclosed. A NUMA computer system includes at least two nodes coupled to an interconnect. Each of the two nodes includes a local system memory. In response to a request for evicting an entry from a sparse directory, an non-intervention writeback request is sent to a node having the modified cache line when the entry is associated with a modified cache line. After the data from the modified cache line has been written back to a local system memory of the node, the entry can then be evicted from the sparse directory. If the entry is associated with a shared line, an invalidation request is sent to all nodes that the directory entry indicates may hold a copy of the line. Once all invalidations have been acknowledged, the entry can be evicted from the sparse directory.

Abstract: A method for avoiding livelocks due to stale exclusive/modified directory entries within a non-uniform memory access (NUMA) computer system is disclosed. A NUMA computer system includes at least two nodes coupled to an interconnect. Each of the two nodes includes a local system memory. In response to an attempt by a processor of a first node to read a cache line at substantially the same time as a processor of a second node attempts to access the same cache line, wherein the cache line has been silently cast out from a cache memory within the second node even though a coherency directory within the node still indicates the cache line is held exclusively in the second node, the processor of the second node is allowed to access the cache line only if the second node is an owning node of the cache line. The processor of the first node is then allowed to access the cache line.

Abstract: A method for avoiding data loss due to cancelled transactions within a non-uniform memory access (NUMA) data processing system is disclosed. A NUMA data processing system includes a node interconnect to which at least a first node and a second node are coupled. The first and the second nodes each includes a local interconnect, a system memory coupled to the local interconnect, and a node controller interposed between the local interconnect and a node interconnect. The node controller detects certain situations which, due to the nature of a NUMA data processing system, can lead to data loss. These situations share the common feature that a node controller ends up with the only copy of a modified cache line and the original transaction that requested the modified cache line may not be issued again with the same tag or may not be issued again at all.

Abstract: A queue includes a data multiplexer having an output and at least two inputs and a plurality of data latches. The data latches include at least a first data latch and a second data latch, which each have a data input and a data output. The data output of the first data latch is coupled to a first input of the data multiplexer, and the output of the data multiplexer is coupled to the data input of the second data latch. A data value to be stored in the queue is received at a second input to the data multiplexer. In response to one or more control signals, the data value is latched into at least one of the first and second data latches, thereby storing the data value in the queue. Depending upon the design of the control logic, the queue can implement either first in, first out (FIFO) or last in, first out (LIFO) behavior.

Abstract: A non-uniform memory access (NUMA) computer system includes at least two nodes coupled by a node interconnect, where at least one of the nodes includes a processor for servicing interrupts. The nodes are partitioned into external interrupt domains so that an external interrupt is always presented to a processor within the external interrupt domain in which the interrupt occurs. Although each external interrupt domain typically includes only a single node, interrupt channeling or interrupt funneling may be implemented to route external interrupts across node boundaries for presentation to a processor. Once presented to a processor, interrupt handling software may then execute on any processor to service the external interrupt. Servicing external interrupts is expedited by reducing the size of the interrupt handler polling chain as compared to prior art methods.

Abstract: A data recirculation apparatus for a data processing system includes at least one output buffer from which data are output onto an interconnect, a plurality of input storage areas from which data are selected for storage within the output buffer, and selection logic that selects data from the plurality of input storage areas for storage within the output buffer. In addition, the data recirculation apparatus includes buffer control logic that, in response to a determination that a particular datum has stalled in the output buffer, causes the particular datum to be removed from the output buffer and stored in one of the plurality of input storage areas. In one embodiment, the recirculated data has a dedicated input storage area.

Abstract: A non-uniform memory access (NUMA) computer system includes first and second processing nodes that are each coupled to a node interconnect. The first processing node includes a system memory and first and second processors that each have a respective one of first and second cache hierarchies, which are coupled for communication by a local interconnect. The second processing node includes at least a system memory and a third processor having a third cache hierarchy. The first cache hierarchy and the third cache hierarchy are permitted to concurrently store an unmodified copy of a particular cache line in a Recent coherency state from which the copy of the particular cache line can be sourced by shared intervention.

Abstract: A non-uniform memory access (NUMA) computer system includes a node interconnect and a plurality of processing nodes that each contain at least one processor, a local interconnect, a local system memory, and a node controller coupled to both a respective local interconnect and the node interconnect. According to the method of the present invention, a communication transaction is transmitted on the node interconnect from a local processing node to a remote processing node. In response to receipt of the communication transaction by the remote processing node, a response including a coherency response field is transmitted on the node interconnect from the remote processing node to the local processing node. In response to receipt of the response at the local processing node, a request is issued on the local interconnect of the local processing node concurrently with a determination of a coherency response indicated by the coherency response field.