Analyzing scalability of parallel systems with unbalanced workload
This paper presents a new formulation of the isoefficiency function which can be applied to parallel systems executing balanced or unbalanced workloads. This new formulation allows analyzing the scalability of parallel systems under either balanced or unbalanced workloads. Finally, the validity of this new metric is evaluated using some synthetic benchmarks. The experimental results allow assessing the importance of considering the unbalanced workloads while analyzing the scalability of parallel systems.
KeywordsScalability analysis Isoefficiency Workload imbalance
- 1.Chen J, Taylor V (1998) Mesh partitioning for distributed systems. In: Proceedings of seventh IEEE international symposium on high performance distributed computing Google Scholar
- 2.Grama A, Gupta A, Karypis G, Kumar V (2003) Introduction to parallel computing, 2nd edn. Addison-Wesley, Reading Google Scholar
- 5.Pastor L, Bosque JL (2001) Efficiency and scalability models for heterogeneous clusters. In: Third IEEE international conference on cluster computing. IEEE Comput Soc, Los Alamitos, pp 427–434 Google Scholar