Proportionate resource utilization based VM allocation method for large scaled datacenters

Abstract

The newly emerged concept of dedicated servers for virtual machines in Cloud Data Centres has motivated researchers’ community to think more critically on optimal utilization of host resources. VMs hosted on dedicated servers are not allowed to migrate during their lifetime, thus it is very important for service providers to adopt that VM allocation policy which utilizes the resources of servers in an optimal way in order to improve the performance of whole data centre. Efficient utilization of resources of data centre is quite challenging due to unpredictability of workload. Host machines in a cloud data centre deal with heterogeneous VMs which have different RAM and CPU requirements. Compute intensive VMs require more computing power than RAM where as data intensive VMs require less computing power than RAM. Similarly other VMs like Small and Micro too have different RAM and computing capacity requirements. An improper VM allocation policy to place different types of VMs in hosts, sometimes leads to uneven usage of host resources and results in wastage of Hosts’ resources like RAM and computing power. In the present work, a novel VM allocation policy has been proposed that allocates VMs to hosts keeping in view the RAM and CPU consumption of host in a proportionate way. It uses the concept of skewness to measure the unevenness in usage of host resources and allocates VM to that host machine which has least skew value. Experimental results obtained through simulation are compared with the Simple First Fit and Power Aware Best Fit Decreasing (PABFD) Policies. It has been found that proportionate usage of RAM and CPU capacity of the host machine accommodates more VMs and reduces the total energy consumption of the data centre. It also outperforms PABFD in terms Energy Consumption and Number of Hosts shutdown.