32 results about "High performance parallel computing" patented technology

The invention pertains to a system and method for dispatching and executing the compute-intensive parts of the workflow for database queries on an attached high-performance, parallel computing platform. The performance overhead for moving the required data and results between the database platform and the high-performance computing platform where the workload is executed is amortized in several ways, for example,by exploiting the fine-grained parallelism and superior hardware performance on the parallel computing platform for speeding up compute-intensive calculations,by using in-memory data structures on the parallel computing platform to cache data sets between a sequence of time-lagged queries on the same data, so that these queries can be processed without further data transfer overheads,by replicating data within the parallel computing platform so that multiple independent queries on the same target data set can be simultaneously processed using independent parallel partitions of the high-performance computing platform.A specific embodiment of this invention was used for deploying a bio-informatics application involving gene and protein sequence matching using the Smith-Waterman algorithm on a database system connected via an Ethernet local area network to a parallel supercomputer.

A system, method and computer program product for supporting system initiated checkpoints in high performance parallel computing systems and storing of checkpoint data to a non-volatile memory storage device. The system and method generates selective control signals to perform checkpointing of system related data in presence of messaging activity associated with a user application running at the node. The checkpointing is initiated by the system such that checkpoint data of a plurality of network nodes may be obtained even in the presence of user applications running on highly parallel computers that include ongoing user messaging activity. In one embodiment, the non-volatile memory is a pluggable flash memory card.

The invention discloses a method for submitting an OpenCL kernel. The method comprises the following steps of 1, transmitting kernel information to the OpenCL kernel; 2, converting the dimension of the work item from one dimension to two dimensions; 3, checking whether the OpenCL kernel uses a local memory or not, and applying for space storage if the OpenCL kernel uses the local memory; 4, checking the size of a local working group; and 5, buffering and submitting the command containing the OpenCL kernel information to hardware equipment. The invention further discloses a device for submitting the OpenCL kernel. The device comprises a kernel information creating and submitting module; a work item dimension conversion module; a local memory variable module; a local work group module; and an OpenCL working thread module. On the premise that the OpenCL standard framework is met, work tasks are reasonably distributed to hardware, the OpenCL kernel is submitted to the hardware to be executed, the parallel operation efficiency is improved, and the method and the device can be widely applied to the field of high-performance parallel calculation.

The present invention provides an OpenFOAM computing task management method based on Web technology. The method is based on Web technology, provides unified, convenient and visualized OpenFOAM cloud computing task management and post-processing visual operation interface and API communication interface, and provides CFD for multi-user usage scenarios. The calculation task creation and management mechanism, users do not need to master the remote login and operation skills of the server, and do not need to understand the OpenFOAM calculation details, which significantly reduces the operation difficulty of the OpenFOAM application in the process of building CFD-aided design; this method also uses the CFD multi-task management program. Personnel on duty avoids waste of resources caused by manual monitoring and untimely error handling; through automatic allocation of computing resources, it avoids resource competition caused by multi-user use, resulting in reduced efficiency; fully utilizes the multi-core high-performance parallel computing capabilities of computing servers to maintain The server can continuously and efficiently carry out CFD analysis of construction projects, providing a new technical means for CFD batch automatic calculation.

The invention provides a Cache simulator based on a GPU and a time parallel speedup simulating method thereof. The Cache simulator comprises an initializing module, a Trace segmentation module, a GPU parallel simulating module, a GPU parallel modifying module and a statistical calculation module. The cache simulator is based on the GPU with strong high-performance parallel computing capacity, hasmulti granularity and multi configuration abilities and parallel simulation characteristic. The invention adopts a time parallel speedup method to segment longer Trace arrays, realizes the parallel simulation of multi Trace segments on the GPU, and utilizes the GPU parallel modifying module to modify the error caused by the modifying and simulating processes in the simulating process. The invention has the advantages that the Cache simulating efficiency is improved, computing resources are better utilized and simultaneously higher cost-performance ratio is achieved.

The invention provides a high-performance parallel computing method based on external PCI-E connection. The method includes the steps of (a) connecting all pieces of computer equipment through a PCI-E bus, (b) running a parallel computing program, (c) constructing a virtual memory addressing table, (d) sending virtual memory information, (e) receiving the virtual memory information, (f) judging correctness of the received information, (g) judging whether a virtual memory address is constructed, (h) allocating computing tasks, (i) executing the computing tasks, and (j) obtaining results and providing the results for a user. The data transmission speed between processors is increased in the parallel computing process through the parallel computing method; the parallel computing method is improved, so that the copy data size of a system CPU is reduced, efficiency of parallel computing operation is greatly improved, multiple parallel computing resources are effectively connected for communication, and data are transmitted at a high speed.

The present disclosure provides a data processing method. By separating and using the HPC storage device and the distributed storage device, when high-performance parallel computing is required, the HPC storage device is used to store the pre-copied data to be processed from the distributed storage device, responsible for Provide parallel computing real-time data for each computing node of the HPC cluster. The computing result data of each computing node is transferred to the distributed storage device. The distributed storage device is used to store the data uploaded by the terminal, the calculation result data and the subsequent data that needs to be processed by the terminal. In this way, the access volume of the HPC storage device and the network bandwidth burden during parallel data calculation can be reduced, the capacity requirement of the HPC storage device can be reduced, and the total cost of ownership of the HPC storage device can be reduced. In addition, when high-performance parallel computing is required, there is no need for the terminal to upload data or download calculation result data separately, which saves data operation time and facilitates the sharing of calculation result data.

The present invention provides a stirred tank reactor simulation method based on an immersed boundary method. The method includes: a preparation phase including determination and setting of calculating parameters, generation of fluid grid files, generation of Lagrange mark point information, and preparation of calculating resources; numerical calculation: setting boundary conditions of computational domains and solving velocity fields and temperature fields based on fluid grids; combining with a discrete element method to achieve an interface analytic simulation of two-phase flow of particle flow, wherein the immersed boundary method is directly employed to apply non-slip boundary conditions to particle surfaces; a post-processing phase, and outputting information of each Euler mesh and information of each particle after a simulation is completed. According to the characteristics of different boundaries, in the calculation, the mixed immersed boundary method is employed to process, and the mass numerical simulation of turbulent flow in the stirred tank reactor is achieved through combination of large eddy simulation and high performance parallel computing technology. The simulation method may be applied to simulation of turbulent flow and simulation of solid particle suspension in the stirred tank reactor and the like.

The invention discloses an MPI (Moldflow Plastics Insight) information scheduling method based on reinforcement learning under a multi-network environment, aiming at overcoming the defect of low practical application performance of a high-performance parallel computer, caused by the traditional circulating scheduling method. The method comprises the steps of: initiating parameters in a process of starting an MPI system, creating Cm<2> Q tables according to a multiple Q table combined method for a computing environment matched with m networks; continuously receiving an MPI information sending request sent by application in a process of starting the MPI system, determining a current information segment, then obtaining a current environment state, scheduling the current information segment to an optimal network according to the state information of historical empirical values stored in the Q tables; and finally, computing an instant reward value obtained by the scheduling and updating Q values in the Q tables. By adopting the invention, the problems that communication loads are distributed unequally, can not adapt to the network state dynamic change and have poor adaptability on the computing environment can be solved, and the practical application performance of the high-performance parallel computer is improved.

A method for optimizing inter-processor communication and cache space utilization in solving structured grid sparse lower triangular equations in high-performance parallel computing, including: receiving the solution vector of structured grid, etc., starting multiprocessors to start parallel execution ;Decompose the solution vector into multiple sub-blocks, divide each sub-block into multiple columns, treat multiprocessors as a rectangular array, each processor is responsible for computing a column, and the processors open up storage dependent data in their own caches Buffer, get the data required for calculation from neighbors or its own memory and store it in the buffer, complete its own column calculation, and send the calculated data to other processors that depend on the data; the rectangular array of processors traverses all sub-blocks, and completes Calculate the required data communication, and perform calculations for each sub-block until the entire solution calculation is completed. The dependent data of each operation is transmitted through fast communication between processors, minimizing memory access overhead and maximizing resource utilization.