VMeasure: Performance Modeling for Live VM Migration Measuring

Abstract

The datacenter is a group of computing resources like networks, servers, storage, etc. These resources provide on-demand access to cloud computing. The multiple instances can run simultaneously using virtualization, and virtual machines (VMs) are being migrated for load balancing, energy optimization, and fault tolerance. When servers are heavily loaded or running large data, migration of a VM from one host to appropriate another host is a must. The performance of live migration is evaluated by getting optimal migration cost. The existing Xen-based migration is designed based on simple techniques such as LRU and compression. On the other hand, a number of techniques have been applied to predict dirty pages while migrating VM. On both techniques including Xen-based and prediction, the lacking of dirty pages monitoring is the key issue which does not handle VM migration properly. In our research work, we have applied exponential model to handle dirty pages efficiently. The proposed model is designed based on keeping maximum WWS on constant dirty rate. The state of the art is shown that migration time taken by number of iterations will be \((W_i)max/R\) (for memory intensive pages) otherwise \((W_i)avg/R\). The experimental results show that the vMeasure approach is able to give optimal downtime and migration time on three different workloads. The proposed model (called vMeasure approach) is able to reduce 13.94% downtime and 11.76% total migration time on an average.