Min-max exclusive virtual machine placement in cloud computing for scientific data environment

Abstract

Abstract In cloud computing, there is a trade-off between SLAV (Service Level Agreement Violation) and system operating cost. Violation rates can be decreased when using more hosts, which increases system operating costs. Therefore, to manage the resources of those hosts efficiently, finding an optimal balancing point between SLAV and system operating cost is critical. In addition, a cost-effective load balancing approach based on the proper migration of VMs (Virtual Machines) in the hosts is needed. For this purpose, some indicators are also necessary to identify the abnormal hosts that violate SLA. One of the primary indicators, CPU usage, is closely related to energy consumption and can be used to reduce SLAV and energy consumption effectively. Our approach focuses on the special environment such as the cloud environment for the scientific data. Here, most of the jobs are data-intensive and a large amount of disk operations is required. Owing to disk operations are likely to affect the performance degradation of the host, disk bandwidth usage as well as CPU usage should be also considered. In this study, we propose the Min-Max Exclusive VM Placement (MMEVMP) strategy to minimize both SLAV and energy consumption. The current working system called KIAF (KISTI Analysis Facility), the CERN ALICE experimental cloud environment for scientific data analysis, is used to analyze the characteristics of data-intensive jobs within it. In this experiment, a lightweight version of the CloudSim simulator was developed and the results were compared to the other methods of different policies. Our evaluation showed that our proposed strategy can reduce SLA violation reasonably as well as the system operating cost-effectively.

Keywords

• Cloud computing
• Scientific data
• Data-intensive
• VM placement
• Disk load balancing