A SLA driven VM auto-scaling method in hybrid cloud environment

The advent of Science Clouds enables scientists to facilitate large-scale scientific computational experiments over cloud environment besides specialized supercomputers in diverse science domains. Cloud computing service elicits efficiency on on-demand resource usage and timely execution at any given time depending on experimental requirements. Hybrid clouds, composing of private and public clouds, even extend research opportunities on resource selection for further complicated experiments but increase the needs of dynamic resource management to maximize its utilization. At existing public cloud providers for commercial use, rule-based and schedule-based mechanisms have been tried for automatic resource allocation to provide resources for processing dynamic workload of modern applications. However, most of the auto-scaling methods just simply support performance metric such as CPU utilization but rarely are aware of Service Level Agreements (SLA) including execution deadline or cost. In this paper, we propose an auto-scaling method that automatically allocates resources depending on variable resource requirements in hybrid clouds satisfying a user's requirements on SLA. We present experimental results which show that the proposed auto-scaling can minimize SLA violations and acceptable cost if needed.