An Empirical Study of VM Provisioning Strategies on IaaS Cloud

Cloud computing technology is rapidly emerging as quite an efficient execution platform for even highly trusted scientific applications. Efficient resource management plays a pivotal role in the execution along with attaining high performance standards in Distributed environments. Resource provisioning and scheduling has become an important area of research in cloud resource management. Clouds can be handled in a more predictable way through a well designed provisioning strategy. Workflow and bag of tasks are the two main categories of applications usually considered for execution in a cloud environment. Bag of task consists of a number of independent tasks whereas workflow is a combination of several mutually dependent tasks. An execution of scientific workflows (more dependencies) will incur more data storage and communication cost. Here we empirically study how the cost and makespan for execution of scientific applications varies with increase in file size and network bandwidth. This paper deals first with the provisioning strategies and henceforth investigates the impact of Virtual Machine (VM) provisioning strategies when different file sizes are used. The role of data and communication being interpreted here as other costs in case of scientific workflow applications has been taken here as a comprehensive study. Our study here in this paper uses simulation to analyze their actual impact practically on the VM provisioning strategy used.