Use of data sampling to improve parallel processing of computations

    公开(公告)号:GB2489562A

    公开(公告)日:2012-10-03

    申请号:GB201204629

    申请日:2012-03-16

    Applicant: IBM

    Abstract: A method and computer program is provided for performing approximate run-ahead computations. A first group of compute engines (330) is selected to execute full computations (305) on a full set of input data (360). A second, preferably smaller, group of compute engines (340) is selected to execute computations which may approximate on a on a sampled subset of the input data. A third group of compute engines (350) is selected to compute a difference in computation results between first computation results generated by the first group of compute engines and second computation results generated by the second group of compute engines. The second group of compute engines is then reconfigured based on the difference generated by the third group of compute engines. Reconfiguration of the second group may be based on the accuracy of the approximate computations and a measure of confidence using available compute engines (370). Selection of the compute engines may be based on performance capabilities, current workloads or physical affinity. Compute engines may be functional units within a processor, processing devices or special purpose accelerators.

Patent Agency Ranking