A framework for proactive resource allocation in IaaS clouds

Abstract

This work proposes a framework for proactive resource allocation in IaaS clouds to improve the system performance. In this system framework, the over-commit approach is applied to improve resource utilization and to show the performance impact. Although the over-commit approach could obtain better resource utilization, there are still some risks. When the amount of aggregated resources requested by virtual machines exceeds that of available resources on the over-committed physical machines, the service response time of some requests may significantly degrade resulting in the violation of service level agreements (SLA). Our experiments show that the over-commit approach indeed improves the resource utilization. However, when the degree of applying the over-commit approach increases, experiments also show that the burden of this approach increases. So, this work proposes a framework of the proactive mechanism to reduce the burden of applying the over-commit approach by our profiling mechanism. This proposed framework could dynamically provide cloud resources for solving the performance bottleneck of big data applications. Experimental results show the performance improvement of our proposed framework in IaaS clouds.