Analyzing the Cost-Effectiveness of Running HPC Applications on Microsoft Azure Virtual Machines
Rawan Aljamal, Fahed Jubair, Ali El-Mousa · 2021
Nowadays, high-performance computing (HPC) is delivered as a cloud service. Microsoft Azure, one of the prominent cloud service providers, has a designated class of virtual machines (VM) that is recommended for HPC programs. This study presents a cost-effectiveness analysis of using the designated VM type, compared to other cheaper VM types on Microsoft Azure. The study analyzes the communication demand of five message-passing (MPI) programs available in the NAS benchmarks suite on Microsoft Azure VMs, and explains the correlation between cost saving and communication behavior in these programs. The findings of this paper imply the need for future research to consider using machine learning approaches for guiding users towards a more cost-effective use of available VM types on Microsoft Azure for running HPC applications.