Performability analysis of web cache server clusters applied to server virtualization

Abstract

The growth of computers processing power have been renewing the interest in virtualization area. Through virtualization technique, a single server has its resources partitioned into multiple separated execution environments, increasing computer resources usage efficiency. To apply server virtualization, it is necessary to select a virtualization approach, which may differ from complete emulation of entire underlying hardware, as full virtualization, to lightweight alternatives, such as container-based virtualization, that partitions the physical machine resources into multiple isolated user-space instances. Regardless of virtualization approach, application of server virtualization drives to increase of system overhead. So, aforementioned efficiency improvement has led to a new challenge: even with additional overhead of virtualization, is it possible to provide a service on virtualized environment with similar, or even better, performance of non-virtualized infrastructures? This dissertation proposes a measurement performance evaluation to quantitatively investigate the behavior of web cache server clusters on virtualized and non-virtualized environments. First analyzed results not only endorsed impact over performance, but also revealed availability degradation on providing virtualized web cache server clusters. To tackle this issue, proposed methodology focused on overhead reduction to reach five 9’s of availability when providing web cache server cluster on virtualized infrastructure. Additionally, applying screening fractional factorial experiment technique resulted in favorable performance metrics on virtualized environments. Results of executed use cases enable to identify scenarios in which virtualized web cache server clusters perform better in comparison with non-virtualized ones, producing recommendations about optimized usage of investigated virtualized clusters.