| Summary: | 碩士 === 國立清華大學 === 資訊工程學系 === 102 === VM consolidation has been shown as a promising technique for saving energy costs and improving resource utilization of a data center. It relies on VM migration to move user tasks onto fewer numbers of physical servers during o peak hour, and then shutdown those idle servers. However, VM migration is a costly operation that could cause several concerns, such as performance degradation, system instability,
etc. Hence many existing algorithms were proposed to minimize the migration cost at runtime after consolidation is triggered by SLA violation or resource utilization
condition. In contrast, this paper aims to pro-actively prevent VM migration in-advanced through a combination of semi-static VM consolidation strategy and runtime VM placement strategy. On one hand, our VM consolidation strategy aims to minimize SLA violation and maximize resource utilization according to the periodic workload pattern. On the other hand, our VM placement strategy attempts to minimize VM migration and communication cost based on the knowledge server turn-on/o time and task execution time. We evaluate our approach using a real
HPC cluster trace as well as a set of synthetic generated workloads. The results show that our VM management approach can signicantly reduce the number of migrations by 37~46% and communication cost by 5~25% while reducing the
energy cost by 21~41% without causing SLA violation.
|
|---|
