A Communication Latency Predict-Based Method for Deploying Applications on Cloud

Abstract

Nowadays, more and more applications are moving to cloud computing. How to deploy these applications optimally is a great challenge. Many cloud applications, such as scientific applications, are large-scale distributed systems that are deployed on a lot of distributed cloud nodes, and there are a lot of communications between these nodes. Therefore, taking the communication latency between cloud nodes into consideration during the deployment will better the performance of the application. However, the communication latency between nodes cannot be obtained in some scenarios, in which the communication latency monitor daemon is crashed in some cloud nodes. In this paper, we propose a communication latency predict-based method for deploying applications on cloud to address this challenge. Our method predicts the missing value of communication latency via collaborative filtering. And then the clustering-based method is applied to deploy applications on cloud based on the complete communication latency information. Comprehensive experiments are conducted by employing a well-known MPI benchmark and comparing the performance of our method with those of other methods. The experimental results show the effectiveness of our method.