Performance Analysis of Amazon Cloud
Nikita Attiguppe DasharathOral Defence Date:
Friday, December 14, 2012 - 17:00Location:
Professors Jozo Dujmovic and Dragutin Petkovic
The shift to the cloud computing paradigm has been accelerated with the recent advances in high speed and low cost interconnection of off-the shelf computational and storage resources. This made the construction of massive datacenters possible. Infrastructure as a Service (IaaS) is a model which delivers cloud computing infrastructure to users to acquire and release resources according to their demand and gives them the ability to pay according to their usage. Some of the key features of IaaS delivery model are dynamic scaling, variable cost, on-demand self service, broad network access, resource pooling and measured service. IaaS is suitable for organizations who do not have the capital to invest in hardware and when the demand is variable and spiky. Such users need to understand the regulatory compliance of cloud providers. Resources in the cloud are provisioned as virtual machines. Often more than one user is assigned the same hardware. While virtualization is regarded as an overhead for workload execution we still need a thorough and empirical investigation of its impact. Although studies examining the policies employed in cloud environments exist, current and potential IaaS users need a deeper insight into the achieved performance and incurred cost of using cloud. In this project the problem was addressed with the use of SpeedMark and DiskMark, processor and disk performance measurement benchmark programs respectively. These programs generate workloads stressing the compute and disk components. With the use of these benchmarks the overheads imposed by the cloud software stack to workload execution were studied. Subsequently the three major cloud providers namely Amazon EC2, Windows Azure and IBM were analyzed by performing empirical experiments on their virtual machines based on virtual machine configuration and geographical location of the data centers.