LiiT
Laboratory for Internet and Innovative Technologies
Abstract

Multilevel cache memory is used to speedup the average memory access of reused data and data locality. This is especially emphasized for cache intensive algorithm such as matrix vector (MV) multiplication algorithm. Different tools and programming languages exist which speedup the execution by using parallel implementation of a certain algorithm. In this paper we analyze the performance of MV multiplication algorithm using sequential and parallel implementations developed in high level language C# with threads for parallel implementation. We also analyze the performance of the algorithm while executed in the same runtime environment, but hosted in Eucalyptus open source cloud. The research goal is to confirm our hypothesis that cloud computing virtualization degrade cache intensive algorithm performance compared to traditional onpremise algorithm execution. Our research methodology assesses algorithm performance represented with speed and speedup in order to compare and identify algorithm performance drawbacks and fluctuations. The results negate the hypothesis and show that virtualization based cloud computing architectures improve the performance of cache intensive algorithm that we setup instead of degrading it, both for sequential and parallel implementations.

Authors

Goran Velkoski, Sasko Ristov, and Marjan Gusev

Keywords

Cloud Computing, Eucalyptus Cloud, Virtualization, HPC, Cache Memory, Speed, Speedup

Full Paper

The technical report is published as a paper in Proceedings of the 10th Conf on Informatics and Information Technologies, (Editors Mishkovski and Ristov) CiiT 2013, Bitola, Macedonia, 2013