Análise do impacto de migração de máquinas virtuais em ambiente computacional virtualizado

Abstract

Virtual machines migration has been widely applied in datacenter environments due to various benefits including load balancing, fault tolerance and energy saving. However, the migration process causes degradation on the services provided by the virtual machines. In this paper, we present an impact evaluation of two virtual machines migration approaches that have been widely cited in the literature: stop-and-copy and pre-copy. We present an indoor experimental testbed for evaluating these two approaches through four performance metrics: total migration time, downtime, response time and throughput. Experimental results show that (i) the stop-and-copy downtime is at least 5 times higher than the pre-copy one, (ii) the total migration time of pre-copy was higher than stop-and-copy for all workloads, (iii) the pre-copy strategy increases the applications response time, while the use of stop-and- copy led to services downtime.