Bounded-Downtime Computation for Virtual Machine Live Migration Based on Memory Alternation Cross Reference

Nowadays, virtualization is an important technique for achieving non-stop and dependable computing. Live migration is one of the core mechanisms for migrating ongoing virtual machines across hosts. The computation performed on the migrating machine proceeds intermittently due to the side effect of live migration. The intermittent computation leads to an adverse impact on the system performance in general. Thus, bounded downtime is a fundamental requirement for many important applications. In this paper, a method which analyses memory access pattern vertically and horizontally is developed for bounded-downtime applications. The method predicts and resizes memory space available to the application such that the computation can proceed in accord with its response time limit. Experiments are also conducted to show that bounded downtime can be achieved with negligible performance variation.