PERFORMANCE EVALUATION OF K-MEANS BALANCING TECHNIQUE IN CLOUD COMPUTING USING CLOUD ANALYST SIMULATOR

Abstract

The issue of load balancing (LB) has remained a significant challenge in achieving desired performance levels of cloud computing (CC). This paper presents an LB approach based on clustering of virtual machines using the K-Means technique based on parameters MIPS and RAM. The proposed clustering based algorithm has been implemented using the Cloud Analyst (CA) simulator and compared with three existing LB techniques namely, Round Robin (RR), Equally Spread Current Execution Load (ESCEL), and Throttled technique. These algorithms have been evaluated under three different service broker policies (SBP) such as Closest Data Center (CDC), Optimize Response Time (ORT), and Dynamic Reconfiguration (DR) using a dataset of six user bases distributed over different regions which includes millions of peak users. The results indicate that the K-Means algorithm is always better than RR and ESCEL under all three SBPs. It has been found that under the DR policy, K-Means utilized DCPT by reducing it by 92.17% as compared to RR and 67.61% as compared to ESCEL with RT being improved by 78.94% as compared to RR. When looking at the functioning of K-Means during ORT and DR, the numbers say that K-Means produces slightly inferior DCPT as compared to Throttled, which suggests that the utilization of resource-profile-aware cluster dispatch is more effective than using concurrency-capping algorithm while allowing broker policy to become more dynamic.