Cloud computing allows elastic deployment of client applications by provisioning virtual machine(VM) instances. In practice, each VM instance is treated as a non-fractional atomic unit in provisioning, and most cloud providers offer a number of different options for VM instances. A natural question here is what are the tradeoffs of these different VM instances? Or from each client's perspective, which VM instance(s) should it choose given its specific requirement? It is necessary to decide whether to scale up by upgrading existing VMs or scale out by adding more VMs. We study a case of video compression computation offloading to the cloud computing and present a systematic experimental study on a state-of-the-art cloud system to understand the performance of diverse VM instances and the tradeoffs therein. Our results reveal the critical performance bottlenecks of these VM instances, particularly in the presence of resource contention, providing valuable guidelines for computation offloading opportunities and the selection of VM instances in practice.
Copyright is held by the author.
The author granted permission for the file to be printed and for the text to be copied and pasted.
Member of collection