An Application Framework for the Scheduling of HPC Jobs on Hybrid Hardware Platforms for Solving Bioinformatical Problems

Outline

H. Brandstaetter-Mueller, A. Hoelzlwimmer, B. Parsapour, G. Lirk, P. Kulczycki - An Application Framework for the Scheduling of HPC Jobs on Hybrid Hardware Platforms for Solving Bioinformatical Problems - 55th GMDS Congress, Mannheim, Germany, 2010, pp. 354-355

Abstract

Managing a group of heterogeneous compute nodes, especially if the heterogeneity concerns the type of main processing unit, is a nontrivial task. This framework sets out to allow the dynamic and intelligent scheduling and distribution of computational tasks across such a hybrid environment. The framework supports dynamic managing of the tasks as well as a prediction of run times for some often-used algorithms, which allows the scheduler to further optimize the selection of compute nodes to run the calculation on.