Dynamic Virtual Machine Allocation to Minimize Task Migrations in Cloud

Abstract

Cloud Clients (CC) in a virtualized Cloud environment have no assurance of the quality on the services which are being offered by the Cloud providers, particularly in context with the deadline of the tasks submitted by the CC. The Service Level Agreements (SLA) is mutually agreed upon by the CC and the Cloud provider which involves the requirement of resources from the CC including CPU, memory and bandwidth. The other services namely the deadline time for the task, response time are also included in the SLA. The allocation algorithms for allocating Virtual Machines (VM) to the tasks have a major role to play in determining the quality of services provided to the CC. The major problem with current allocation algorithms includes the decisions as to when the tasks need to be migrated from one VM to another, large waiting time for the task execution and not meeting the deadline threshold. This paper addresses dynamic allocation issue for allocating VM to tasks. The proposed approach of allocating VM to task is split into two phases. In the first phase, an algorithm is proposed for dynamic allocation of VM to tasks. In the second phase, another algorithm is proposed for determining whether the migration of task is needed or not. The proposed approach significantly reduced the waiting time, completion time and idle time. The proposed algorithm results are compared with Max-Min and Min-Min approaches and the results shows 15 percent reduction in waiting time, 11 percent reduction in completion time and 13 percent reduction in idle time of VMs compared to the Min-Min approach and 12 percent reduction in waiting time, 10 percent reduction in completion time and 11 percent reduction in idle time of VMs compared to Max-Min approach.