Workload management in a distributed processing environment

    公开(公告)号:GB2357863A

    公开(公告)日:2001-07-04

    申请号:GB0019541

    申请日:2000-08-10

    Applicant: IBM

    Abstract: Upon receipt of a request for service, 202, a workload manager according to the present invention classifies the request, 204, into a particular type of service and places the work on a logical queue to be executed, 206. Routing queues place the work, using initiators, 208, on one of many available nodes. The work is distributed to the nodes based on accounting, availability and past performance of all nodes considered. Once the work is completed notification is given to the management module, 216. Changes are made periodically to the system to meet demands for service and/or conserve resources, 212, in accordance with recorded statistics. These changes are made by a policy manager which may start additional initiators or add additional router capacity to handle the current workload. The workload manager learns the need of programs over time, 218, and does not require the intervention of a system administrator. The system does not rely on system specific metrics but may use performance metrics. Additionally, the workload manager allows objects to be distributed through the cluster of workstations, while behaving as if they were initiated locally on the machine on which they run. The similar type of work sent to different workstations may have the same class names, and the classpath may hold different implementations of that object at a particular workstation. Thus, at dynamic compilation time, if a platform has a more efficient feature than others, a work unit is dynamically compiled with the more efficient feature.

Patent Agency Ranking