APPARATUS AND METHOD FOR EXECUTING MAINTENANCE PROCESSING ON COMPUTERS COUPLED VIA A MULTIDIMENSIONAL MESH OR TORUS CONNECTION IN A NETWORK

摘要

Computers are connected via multidimensional mesh or torus connection in a network. In response to a request for executing maintenance processing on computers in the network, an apparatus detects execution-scheduled jobs to be executed after an execution start time of the maintenance processing, based on execution-scheduled start times of jobs to be executed by the computers. The apparatus calculates, for each execution-scheduled job, a characteristic value of an axial length of an execution-scheduled job area in each axial direction of multidimensional axes in the network, where the execution-scheduled job area includes a group of computers to execute the each execution-scheduled job. The apparatus determines a maintenance area in the network on which the maintenance processing is to be executed, based on the characteristic values of the axial lengths of the execution-scheduled job areas, and executes the maintenance processing on computers in the maintenance area.

主权项

1. A parallel computer system comprising:
a plurality of computers coupled via multidimensional mesh or torus connection in a network; and a maintenance device configured to: in response to a request for executing maintenance processing on computers in the network, detect one or more execution-scheduled jobs to be executed after an execution start time of the maintenance processing, based on start time information indicating an execution-scheduled start time of a job to be executed by two or more computers, calculate, for each of the one or more execution-scheduled jobs, a characteristic value of an axial length of an execution-scheduled job area in each axial direction of a plurality of multidimensional axes in the network, based on area information indicating a length of an execution-scheduled job area in each axial direction of the plurality of multidimensional axes, the execution-scheduled job area including a group of computers to execute the each execution-scheduled job, determine a maintenance area in the network on which the maintenance processing is to be executed, based on the characteristic values of the axial lengths of the execution-scheduled job areas for the one or more execution-scheduled jobs, and execute the maintenance processing on computers in the maintenance area.