Abstract: A remote translation mechanism for a multi-node system. One embodiment of the invention provides a method for remotely translating a virtual memory address into a physical memory address in a multi-node system. The method includes providing the virtual memory address at a source node, determining that the virtual memory address is to be sent to a remote node, sending the virtual memory address to the remote node, and translating the virtual memory address on the remote node into a physical memory address using a remote-translation table (RTT). The RTT contains translation information for an entire virtual memory address space associated with the remote node.

Abstract: A method and apparatus for a coherence mechanism that supports a distributed memory programming model in which processors each maintain their own memory area, and communicate data between them. A hierarchical programming model is supported, which uses distributed memory semantics on top of shared memory nodes. Coherence is maintained globally, but caching is restricted to a local region of the machine (a “node” or “caching domain”). A directory cache is held in an on-chip cache and is multi-banked, allowing very high transaction throughput. Directory associativity allows the directory cache to map contents of all caches concurrently. References off node are converted to non-allocating references, allowing the same access mechanism (a regular load or store) to be used for both for intra-node and extra-node references.

Abstract: A remote translation mechanism for a multi-node system. One embodiment of the invention provides a method for remotely translating a virtual memory address into a physical memory address in a multi-node system. The method includes providing the virtual memory address at a source node, determining that the virtual memory address is to be sent to a remote node, sending the virtual memory address to the remote node, and translating the virtual memory address on the remote node into a physical memory address using a remote-translation table (RTT). The RTT contains translation information for an entire virtual memory address space associated with the remote node.

Abstract: A cache coherence system and method for use in a multiprocessor computer system having a plurality of processor nodes, a memory and an interconnect network connecting the plurality of processor nodes to the memory. Each processor node includes one or more processors. The memory includes a plurality of lines and a cache coherence directory structure having a plurality of directory structure entries. Each of the directory structure entries is associated with one of the plurality of lines and each directory structure entry includes processor pointer information, expressed as a set of bit vectors, indicating the processors that have cached copies of lines in memory. Processor pointer information may be a function of a processor number assigned to each processor; the processor number may be expressed as a function of a first set of bits and a second set of bits which are respectively mapped into first and second bit vectors of the n bit vectors.

Abstract: A method of node translation for communicating over virtual channels in a clustered multiprocessor system using connection descriptors (CDs), which specify the endpoint nodes for virtual connections. The system includes a local processing element node, a remote processing element node and a network interconnect therebetween for sending communications between the processing element nodes. The method includes assigning a CD to specify an endpoint node for a virtual connection, defining a local connection table (LCT) to be accessed with the CD to produce a system node identifier (SNID) of the endpoint node, generating a communication request including the CD, accessing the LCT using the CD of that communication request to produce the SNID for the endpoint node of the connection in response to that request, and sending a memory request to the endpoint node.

Abstract: A method of performing remote address translation in a multiprocessor system includes determining a connection descriptor and a virtual address at a local node, accessing a local connection table at the local node using the connection descriptor to produce a system node identifier for a remote node and a remote address space number, communicating the virtual address and remote address space number to the remote node, and translating the virtual address to a physical address at the remote node (qualified by the remote address space number). A user process running at the local node provides the connection descriptor and virtual address. The translation is performed by matching the virtual address and remote address space number with an entry of a translation-lookaside buffer (TLB) at the remote node. Performing the translation at the remote node reduces the amount of translation information needed at the local node for remote memory accesses.

Abstract: A multiple counter-rotating ring computer network system having a permission control scheme for client isolation. The peripheral channel allows two rings to be folded into one longer ring so that faulty nodes can be effectively removed from the network. Or, any of the rings can be masked so that they are unoperational. The network system also allows several client isolation states ranging from complete isolation to master access. These types of isolation allow faulty client devices to be tested while maintaining a high-level of network security by configuring the client to an appropriate isolation state.

Abstract: Improved method and apparatus for facilitating fast barrier synchronization in a parallel-processing system. A single input signal and a single output signal, and a single bit of state (“barrier_bit”) is added to each processor to support a barrier. The input and output signal are coupled to a dedicated barrier-logic circuit that includes memory-mapped bit-vector registers to track the “participating” processors and the “joined” processors for the barrier. A “bjoin” instruction executed in a processor causes a pulse to be sent on the output signal, which in turn causes that processor's bit in the dedicated barrier-logic circuit's “joined” register to be set.

Abstract: Improved method and apparatus for facilitating barrier and eureka synchronization in a massively parallel processing system. The present barrier/eureka synchronization mechanism provides a partitionable, low-latency, immediately reusable, robust mechanism which can operate on a physical data-communications network and can be used to alert all processor entities (PEs) in a partition when all of the PEs in that partition have reached a designated barrier point in their individual program code, or when any one of the PEs in that partition has reached a designated eureka point in its individual program code, or when either the barrier or eureka requirements have been satisfied, which ever comes first. Multiple overlapping synchronization partitions are available simultaneously through the use of a plurality of parallel synchronization contexts.

Abstract: A system and method of accessing a memory location within a system having a processor and a plurality of memory locations separate from the processor. The system includes a plurality of external registers which are connected to the processor over a data bus, address translation means, connected to the processor over the data bus and an address bus, for calculating, based on an index written to the data bus, an address associated with one of the memory locations, and transfer means, connected to the plurality of external registers, for transferring data between the addressed memory location and one of the external registers.

Abstract: A congestion control mechanism for a node of a modular computer network system. The mechanism includes registers for maintaining the number of undelivered requests and unanswered requests for the node and registers for the maximum number of such undelivered requests and unanswered requests. The mechanism regulates congestion on the network by throttling back or ratcheting up the allowed number of undelivered requests and unanswered requests based upon the level of busy and non-busy results of such requests and answers. Congestion is also alleviated by the implementation of a set of large and small send and receive buffers. These buffers are configurably partitioned among virtual I/O channels. Each request virtual I/O channel may utilitize congestion control.

Abstract: A system and method of transferring information between a peripheral device and an MPP system having an interconnect network and a plurality of processing nodes. Each processing element includes a processor, local memory and a router circuit connected to the interconnect network, the processor and the local memory. Each router circuit includes means for transferring data between the processor and the interconnect network and means for transferring data between the local memory and the interconnect network. An I/O controller is connected to a plurality of the router circuits. Data is then read from the peripheral device and transferred through the I/O controller to local memory of one of the processing elements.

Abstract: A messaging facility in a multiprocessor computer system includes assembly circuitry in a source processing element for assembling a message to be sent from the source processing element to a destination processing element based on information provided from a processor in the source processing element. A network router transmits the assembled message from the source processing element to the destination processing element via an interconnect network. A message queue in a local memory of the destination processing element stores the transmitted message. A control word stored in the local memory of the destination processing element includes a limit field designating a size of the message queue and a tail field designating an index into the corresponding message queue to indicate a location in the message queue where the transmitted message is to be stored. Shell circuitry in the destination processing element atomically reads and updates the tail field.

Abstract: A system and method of accessing a memory location within a system having a processor and a plurality of memory locations separate from the processor. The system includes a plurality of external registers which are connected to the processor over a data bus, address translation means, connected to the processor over the data bus and an address bus, for calculating, based on an index written to the data bus, an address associated with one of the memory locations, and transfer means, connected to the plurality of external registers, for transferring data between the addressed memory location and one of the external registers.

Abstract: Method and apparatus for a filtered stream buffer coupled to a memory and a processor, and operating to prefetch data from the memory. The filtered stream buffer includes a cache block storage area and a filter controller. The filter controller determines whether a pattern of references has a predetermined relationship, and if so, prefetches stream data into the cache block storage area. Such stream data prefetches are particularly useful in vector processing computers, where once the processor starts to fetch a vector, the addresses of future fetches can be predicted based in the pattern of past fetches. According to various aspects of the present invention, the filtered stream buffer further includes a history table, a validity indicator which is associated with the cache block storage area and indicates which cache blocks, if any, are valid.

Abstract: A congestion control mechanism for a node of a modular computer network system. The mechanism includes registers for maintaining the number of undelivered requests and unanswered requests for the node and registers for the maximum number of such undelivered requests and unanswered requests. The mechanism regulates congestion on the network by throttling back or ratcheting up the allowed number of undelivered requests and unanswered requests based upon the level of busy and non-busy results of such requests and answers. Congestion is also alleviated by the implementation of a set of large and small send and receive buffers. These buffers are configurably partitioned among virtual I/O channels. Each request virtual I/O channel may utilitize congestion control.

Abstract: A routing mechanism includes two acyclic non-adaptive virtual channels having two types of virtual channel buffers to store packets along deterministic virtual paths between nodes in an n-dimensional networked system, and an adaptive virtual channel having a third type of virtual channel buffer to store the packets along non-deterministic virtual paths between the nodes. The packets are routed between the nodes along either selected portions of the deterministic virtual paths or selected portions of the non-deterministic virtual paths based on routing information such that a packet is never routed on a selected portion of one of the non-deterministic virtual paths unless the third type virtual channel buffer associated with the selected portion of the one non-deterministic virtual path has sufficient space available to store the entire packet.

Abstract: A method optimizes routing in a multiprocessor computer system by defining two types of virtual channels having virtual channel buffers for storing messages communicated between processing element nodes in the multiprocessor computer system. A dateline is associated to each type of virtual channel, and messages are restrained from crossing a dateline on its associated type of virtual channel to avoid deadlock. A cost function is defined which is correlated to imbalances in the utilization of the two types of virtual channels. The unrestrained messages are allocated between the two types of virtual channels to minimize the cost function by defining an initial virtual channel allocation, randomly modifying the virtual channel allocation, and accepting the random modification if the modification decreases the cost function, else accepting the modification based on a probability that slowly decreases during the allocating step.