Local Robustness: A Process Migration Criterion in HPC Clusters
Cluster computing systems require managing their resources and running processes dynamically in an efficient manner. Preemptive process migration is such a mechanism that tries to improve the overall performance of a cluster system running independent processes. In this paper, we show that blind migration of processes at runtime by such a mechanism does not lead to better performance. Instead, the preemptive process migration mechanism requires a criterion to determine if the migration of a process would enhance the cluster performance or not. We introduce a criterion called local robustness to guide the mechanism in this respect. The results of our experiments on a real implementation of a mechanism using this criterion have shown improvements to the overall performance of a Mosix cluster in terms of system response time compared to when processes were migrated blindly.
KeywordsCluster Computing Load Distribution Process Migration Local Robustness Criteria
Unable to display preview. Download preview PDF.
- 1.Sterling, T.: Beowulf Cluster Computing with Linux, 2nd edn. The MIT Press, London (2002)Google Scholar
- 2.Sharma, S., Singh, S., Sharma, M.: Performance Analysis of Load Balancing Algorithms. In: 38th World Academy of Science, Engineering and Technology, pp. 269–273 (2008)Google Scholar
- 4.Barak, A., Braverman, A., Gilderman, I., Laadan, O.: The MOSIX Multicomputer Operating System for Scalable NOW and its Dynamic Resource Sharing Algorithms. Technical report, The Hebrew University (1996)Google Scholar
- 9.Online Oxford Dictionary, http://oxforddictionaries.com (last access 2011)
- 11.Khan, Z., Singh, R., Alam, J., Kumar, R.: Performance analysis of Dynamic load Balancing Techniques for Parallel and Distributed Systems. International Journal of Computer and Network Security 2, 123–127 (2010)Google Scholar
- 12.Ghiasvand, S., Khaneghah, E.M., Khorandi, S.M., Mirtaheri, S.L., Nezhad, N.O., Mohammadkhani, M., Sharifi, M.: An Analysis of MOSIX Load Balancing Capabilities. In: International Conference on Advanced Engineering Computing and Applications in Sciences, Lisbon, Portugal, November 20-25 (2011)Google Scholar