Title:HEET: A Heterogeneity Measure to Quantify the Difference across Distributed Computing Systems

Abstract:Although system heterogeneity has been extensively studied in the past, there is yet to be a study on measuring the impact of heterogeneity on system performance. For this purpose, we propose a heterogeneity measure that can characterize the impact of the heterogeneity of a system on its performance behavior in terms of throughput or makespan. We develop a mathematical model to characterize a heterogeneous system in terms of its task and machine heterogeneity dimensions and then reduce it to a single value, called Homogeneous Equivalent Execution Time (HEET), which represents the execution time behavior of the entire system. We used AWS EC2 instances to implement a real-world machine learning inference system. Performance evaluation of the HEET score across different heterogeneous system configurations demonstrates that HEET can accurately characterize the performance behavior of these systems. In particular, the results show that our proposed method is capable of predicting the true makespan of heterogeneous systems without online evaluations with an average precision of 84%. This heterogeneity measure is instrumental for solution architects to configure their systems proactively to be sufficiently heterogeneous to meet their desired performance objectives.