36 results about "Collective communication" patented technology

A system and method for enabling high-speed, low-latency global collective communications among interconnected processing nodes. The global collective network optimally enables collective reduction operations to be performed during parallel algorithm operations executing in a computer structure having a plurality of the interconnected processing nodes. Router devices ate included that interconnect the nodes of the network via links to facilitate performance of low-latency global processing operations at nodes of the virtual network and class structures. The global collective network may be configured to provide global barrier and interrupt functionality in asynchronous or synchronized manner. When implemented in a massively-parallel supercomputing structure, the global collective network is physically and logically partitionable according to needs of a processing algorithm.

Intra-node data transfer in collective communications is facilitated. A memory object of one task of a collective communication is concurrently attached to the address spaces of a plurality of other tasks of the communication. Those tasks that attach the memory object can access the memory object as if it was their own. Data can be directly written into or read from an application data structure of the memory object by the attaching tasks without copying the data to / from shared memory.

A system and method for supporting collective communications on a plurality of processors that use different parallel programming paradigms, in one aspect, may comprise a schedule defining one or more tasks in a collective operation, an executor that executes the task, a multisend module to perform one or more data transfer functions associated with the tasks, and a connection manager that controls one or more connections and identifies an available connection. The multisend module uses the available connection in performing the one or more data transfer functions. A plurality of processors that use different parallel programming paradigms can use a common implementation of the schedule module, the executor module, the connection manager and the multisend module via a language adaptor specific to a parallel programming paradigm implemented on a processor.

A system and method for enabling high-speed, low-latency global collective communications among interconnected processing nodes. The global collective network optimally enables collective reduction operations to be performed during parallel algorithm operations executing in a computer structure having a plurality of the interconnected processing nodes. Router devices are included that interconnect the nodes of the network via links to facilitate performance of low-latency global processing operations at nodes of the virtual network. The global collective network may be configured to provide global barrier and interrupt functionality in asynchronous or synchronized manner. When implemented in a massively-parallel supercomputing structure, the global collective network is physically and logically partitionable according to needs of a processing algorithm.

A collective communication apparatus and method for parallel computing systems. For example, one embodiment of an apparatus comprises a plurality of processor elements (PEs); collective interconnect logic to dynamically form a virtual collective interconnect (VCI) between the PEs at runtime without global communication among all of the PEs, the VCI defining a logical topology between the PEs in which each PE is directly communicatively coupled to a only a subset of the remaining PEs; and execution logic to execute collective operations across the PEs, wherein one or more of the PEs receive first results from a first portion of the subset of the remaining PEs, perform a portion of the collective operations, and provide second results to a second portion of the subset of the remaining PEs.

The invention provides a novel method for optimizing an MPI (message passing interface) application in multi-core environments, and particularly relates to a machine learning based method for predicating parameters during MPI application optimal operation under multi-core clusters. According to the method, training benchmarks with different ratios of point-to-point communication data to collective communication data are designed to generate training data under the specific multi-core clusters, parameter optimized models during operation are constructed by a decision tree REPTree capable of quickly outputting results and an ANN (artificial neural network) capable of generating multiple output and good in noise immunity, the optimized models are trained by the training data generated by the training benchmarks, and the trained models are used for predicating the unknown parameters inputted to the MPI application during optimal operation. Experiments show that speed-up ratios generated by the parameters obtained by the REPTree-based predication model and the ANN-based predication model during optimal operation are averagely higher than 90% of a practical maximum speed-up ratio.

Methods, apparatuses, and computer program products for optimizing collective communications within a parallel computer comprising a plurality of hardware threads for executing software threads of a parallel application are provided. Embodiments include a processor of a parallel computer determining for each software thread, an affinity of the software thread to a particular hardware thread. Each affinity indicates an assignment of a software thread to a particular hardware thread. The processor also generates one or more affinity domains based on the affinities of the software threads. Embodiments also include a processor generating, for each affinity domain, a topology of the affinity domain based on the affinities of the software threads to the hardware threads. According to embodiments of the present application, a processor also performs, based on the generated topologies of the affinity domains, a collective operation on one or more software threads.

A method, system, method and computer program product for facilitating collective communication in parallel computing. A system for parallel computing includes one or more communicators. Each of the one or more communicators comprises a plurality of processes. A memory pool including one or more early arrival buffers is provided. One or more tokens are assigned to a specified communicator included in the communicators. Each of the processes comprised by the specified communicator may consume any token assigned to the specified communicator. Requesting an early arrival buffer included in the memory pool requires consuming at least one token. A collective communication operation is performed using the specified communicator. The collective communication operation is performed eagerly using early arrival buffers obtained by consuming the tokens assigned to the communicator.

System and method for suffix tree creation for large input data / text streams. The methodology leverages the structure of suffix trees to build a suffix tree by simultaneously tiling accesses to both the input string as well as the partially constructed suffix tree. The end result enables the indexing of very large input strings and at the same time maintain a bounded working set size and a fixed memory footprint. The method is employed for serial processing. Further, a scalable parallel suffix tree construction is realized that is suitable for implementation on parallel distributed memory systems that use effective collective communication and in-network caching. The methodology is also applied for suffix link recovery in both serial and parallel implementations.

Parallel application checkpoint image compression may be carried out in a parallel computer. The parallel computer may include a plurality of compute nodes, where each node is configured to execute one or more parallel tasks of the parallel application. The parallel tasks may be organized into an operational group for collective communications. In such a parallel computer, checkpoint image compression may include: generating, by each task of the parallel application, an image for checkpointing the parallel application; selecting, by an image management task, one of the images as a base template image; constructing, by the image management task, a binary radix tree, including storing differences between each task's image and the base template image in the binary radix tree; and storing, by the image management task as a checkpoint for the parallel application, the binary radix tree and the base template image, without storing every task's image.

The invention discloses a collective communication mobile terminal system and application of the system. The system comprises a mobile terminal instant messaging client, a UST module, a UPUSH engine module, and a background server, wherein the mobile terminal instant messaging client runs on the mobile terminal of a user and is used for inputting and displaying information; the UST module is used for verifying a user, managing and storing user information, such as the related account information of the user-bound instant messaging tool and a social website; the UPUSH engine module is used for pushing the communication messages of the mobile terminal instant messaging client to the background server or pushing the messages of the background server to the mobile terminal instant messaging client; and the background server is used for communicating with the user-bound instant messaging tool and the social website. According to the invention, the system can bind a plurality of instant messaging tools to realize unified management of the instant messaging tools and to realize message sending cross the messaging tools.

The invention relates to a collective communication function modeling method of a parallel program, and aims to eliminate the defects in the prior art that communication time data can not be accurately obtained and a great quantity of time and money is consumed. The collective communication function modeling method of the parallel program comprises the following specific process: S1: measuring a collective communication function under an experimental platform for N times, and obtaining the communication time data of the collective communication function under different degrees of parallelism and data sizes; and S2: utilizing an artificial neural network based on a BP (Back Propagation) algorithm to fit the communication time data of the collective communication function under different degrees of parallelism and data sizes to obtain a neural network model of a corresponding communication function. The collective communication function modeling method is used for the field of the communication technology.

A communication method for collective operations is provided which eliminates the need for a three layered structure. Instead, the method of the present invention employs a control structure which is specifically designed to keep track of the progress of the collective communication operations and the facilities for the handling of asynchronous message passing. The method of the present invention thus eliminates the need for an intermediary, point-to-point communication layer that is required to follow message matching semantics for sending messages, receiving messages and for awaiting message arrivals. Application layer tasks and user defined programs running in the communication layer share access to the control structure as a mechanism for synchronizing and scheduling the progress of the collective communication operation.