A Greedy Events Scheduling Method for Network-aware Virtual Machine Placement

• Main Authors: Yu-Chen Hsiao, 蕭宇辰
• Other Authors: 林偉
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/11982266270995883689

碩士 === 國立中興大學 === 資訊科學與工程學系所 === 100 === Advances in virtualization techniques have accelerated the development of cloud computing technology. Connected by high-speed switches, tens of thousands of servers form a datacenter on a warehouse scale to service Internet users with diverse requests: search, social networking, online maps, and many others. The tremendous popularity of such Internet services necessitates efficient resource management and virtual machine placement to keep up with the rapid resource demands. The placement problem of virtual machines has become a focal point of research. However, the existing placement methods emphasize resource optimization, neglecting the effect of networks. This could lead to a dramatic increase in access time for applications to retrieve files or data. In this thesis, we propose a scheduling method for virtual machine placement. The proposed method uses a greedy approach to scheduling tasks on virtual machines. While determining the task schedule, the proposed method considers the minimization of computing time and data access time as well. Virtual machines are checked to make sure that they provide adequate resources, and task completion times over different virtual machines are evaluated. In addition, our method takes data access time into account. The proposed method intends to choose virtual machines in order that data relevant to the applications can be rapidly accessed with minimal network latency. In other words, while determining the placement, our method is aware of the cost of data access through networks. The proposed method conducts a search for the task with the best placement that, for the time being, requires the minimum time of computing and data access combined. And the task is scheduled with the placement. This process repeats until all the tasks are scheduled and their placements are determined. Our proposed method can effectively minimize the completion time of tasks. The simulation results indicate that our method outperforms the conventional round robin(RR) scheduler. The average task completion time becomes smaller by the proposed method than by the RR scheduler, particularly when the number of tasks increases. For example, as the number of tasks approaches 25-30, the average completion time by the RR scheduler shoots up drastically. However, our method can effectively control the increase in the average completion time. And it maintains lower task completion times than the RR scheduler, by a wide margin of 600 seconds.