Title

Investigation of Downey model for speedup prediction

Date of Award

2007

Degree Type

Thesis

Department

Computer Science

Rights

CC BY-NC-ND 4.0

Abstract

In parallel computing, accurate prediction of speedup is important for job schedulers with adaptive resource allocation. The predicted speedup determines the expected runtime on a certain number of nodes and the efficiency by which the resources are used. Among the existing speedup prediction models, the Downey model [5, 6] is simple but promising. However, the prediction accuracy of the Downey model needs to be investigated in realistic scenario setups. In this thesis, we use the NAS benchmarks and synthetic benchmarks [19] to generate scenarios in which the performance of the Downey model is examined. Based on these experiments, conditions are suggested for the successful application of the Downey model.