Two-Stage metascheduling of grid workflows
MetadataShow full item record
Grid meta-schedulers play a key role in computing grid environments. They match grid application jobs with available resources, in order to achieve optimal execution. However, a typical grid application is not an isolated single grid job, but a composition of multiple grid services that require coordinated scheduling and execution. Thus, scheduling grid workflows in grid environments require a dynamic multi-stage scheduling framework. This dissertation proposes a two-stage metascheduling architecture for complex grid workflow applications that decouples logical task meta-scheduling from physical task/node matchmaking. This approach aims at writing grid application workflows with no knowledge of the execution grid infrastructure or topology. It also factures the dynamic aspect of grids at different stages of application execution. To validate this architecture, we built TSM-SIM, a two-stage metascheduler simulator for grid workflow applications. It supports dynamic grid resource and job simulation, and provides a submission interface for workflow grid applications as a single unit, rather than as a set of grid jobs. We present the overall architecture TSM-SIM as well as its scheduling algorithms. We demonstrate how it can be used to collect performance scheduling data of complex grid workflow benchmarks.