HPCCloud Seer: A Performance Model Based Predictor for Parallel Applications on the Cloud

Abdallah Saad,Ahmed El-Mahdy

doi:10.1109/access.2020.2992880

Abstract

With the continual increase in the high performance computing (HPC) market share, the need for a cheaper and widely available system rather than the expensive typical HPC systems increases. A promising alternative to HPC typical systems is the cloud computing environment which is characterised by being cheap, flexible, scalable and available. However, the cloud is based on virtualization which increases the latency to access the processing and network resources due to resource sharing. This makes the cloud an unpredictable environment to long run time programs such as HPC applications. Hence, modelling and understanding performance is essential for exploiting such environment. In this paper we propose a predictor for the execution time of the message passing interface (MPI) based applications on the cloud, as they are a major class of HPC applications. The predictor is based on an analytical performance model through considering the cloud resources as a queueing network, and the parallel applications as jobs contesting for the shared resources. The prediction based on the proposed model is measured on both a cluster of bare-metal servers and on a group of virtual machines. The overall accuracy of this prediction is 88% for 10 benchmarks, 5 from SPEC-MPI and 5 from NASA parallel benchmarks.

Highlights

The interesting features of cloud computing such as availability, elasticity, usability and the ‘pay-as-you-go’ business model, make it suitable for a wide spectrum of applications
EXPERIMENTS SETUP To examine the accuracy of the prediction based on the proposed model and to illustrate the model capabilities, a cluster of virtual machines hosted on a private cloud and a heterogeneous high performance computing (HPC)-cluster have been built sharing the same hardware resources
WORK In this paper we propose a predictor of the execution time of the tightly coupled HPC applications on the cloud

Summary

INTRODUCTION

The interesting features of cloud computing such as availability, elasticity, usability and the ‘pay-as-you-go’ business model, make it suitable for a wide spectrum of applications. The high network latency lessens the scalability of tightly coupled parallel applications; where the cloud unpredictable environment makes it harder for both the customer and the cloud service provider to estimate the cost of running parallel applications on the cloud These obstacles influenced Egwutuoha et al [5] to recommend using bare-metal servers instead of virtual machines in case of tightly coupled applications. This knowledge allows the cloud service providers to afford more flexible and profitable business models for their HPC customers This paper proposes an analytical performance model that predicts the running time of HPC applications on the cloud, either on the cloud VMs or on the cloud bare-metal multicore servers.

RELATED WORK

SERVICE TIME CALCULATION

VISIT RATIO CALCULATIONS

CONCLUSION AND FUTURE WORK