A.C. Sodan and G. Gupta (Canada)
middleware, grid scheduling, adaptation, dynamic resource allocation, job scheduling
Grid applications that run simultaneously on different sites may face varying resource availability on the different sites during their execution. Furthermore, the individual sites are likely to be heterogeneous and potentially apply heterogeneous resource sharing approaches, including time sharing. This makes correct runtime estimation for proper workload distribution difficult. To ensure balanced execution progress of the parallel grid job, the application may, therefore, have to adapt its workload to varying resource availability or to corrections of runtime estimates. We present a unified approach for application internal workload adaptation under different resource sharing types; and an integration of the application internal scheduler with the batch job schedulers per site.
Important Links:
Go Back
