发明名称 Thread scheduling across heterogeneous processing elements with resource mapping
摘要 A system and program product for scheduling processes of a workload on a plurality of hardware threads configured in a plurality of processing elements of a multithreading parallel computing system for processing thereby. Process dimensions for each process are determined based on processing attributes associated with each process, and a place and route algorithm is utilized to map the processes to a processor space representative of the processing resources of the computing system based at least in part on the process dimensions to thereby distribute the processes of the workload.
申请公布号 US9600346(B2) 申请公布日期 2017.03.21
申请号 US201313938683 申请日期 2013.07.10
申请人 International Business Machines Corporation 发明人 Kuesel Jamie R.;Kupferschmidt Mark G.;Schardt Paul E.;Shearer Robert A.
分类号 G06F9/46;G06F9/50;G06F9/48 主分类号 G06F9/46
代理机构 Middleton Reutlinger 代理人 Middleton Reutlinger
主权项 1. A computing system, comprising: a processor; a memory; program code stored in the memory and configured to be executed by the processor to cause the processor to schedule a plurality of processes of a workload on hardware threads of a plurality of processing elements using a place and route algorithm by: representing processing resources of the computing system by generating a processor space having locations corresponding to the hardware threads of the plurality of processing elements, the processor space generated based at least in part on hardware resources associated with the hardware threads of the plurality of processing elements such that hardware threads associated with similar hardware resources are assigned proximate locations of the processor space;assigning process dimensions to the plurality of processes of the workload based at least in part on at least one processing attribute associated with each process among the plurality of processes, the process dimensions representative of one or more locations in the processor space and depending at least in part on the at least one processing attribute associated with each process;analyzing the process dimensions assigned to each process;grouping the plurality of processes into one or more groups based at least in part on the determined process dimensions for each process to map related processes to proximate hardware threads and/or map processes having similar hardware resource requirements to hardware threads in different processing elements;mapping each process among the plurality of processes to at least one location in the processor space based at least in part on the assigned process dimensions using the place and route algorithm; andcausing the processor to schedule each process among the plurality of processes to at least one hardware thread corresponding to the at least one location to which the process is mapped.
地址 Armonk NY US