Interacting SRN Models for Availability Evaluation of VM Migration as Rejuvenation on a System under Varying Workload

摘要

This paper presents a set of analytical models for availability evaluation of a virtualized system with VMM software rejuvenation enabled by VM migration schedule. The proposed models consider aspects of varying workload. We adopted the interacting models approach to reduce time and cost for models' evaluation. Our metrics of interest are steady-state availability and annual downtime. In our first case study, we compare the monolithic model and interacting models approach. Obtained results show a significant reduction in time for model evaluation in the interacting models approach. Our second case study presents a variety of scenarios with a varying workload. We consider the workload variation with two stages: i) peak - when the workload submitted is high; and ii) off-peak - when the workload submitted is low. The evaluation results comprise from systems without high workload demand (peakDuration = 0h per day) to systems with only high workload demand (peakDuration = 24h per day). For all considered scenarios, we present the best rejuvenation schedule to maximize system availability. In scenarios with a heavy workload, the downtime avoidance caused by software rejuvenation surpass three days.