Caching VM Instances for Fast VM Provisioning: A Comparative Evaluation

One of the key metrics of performance in infrastructure cloud is the speed of provisioning a virtual machine (or a virtual appliance) on request. A VM is instantiated from an image file stored in the image repository. Since the image files are large, often GigaBytes in size, transfer of the file from repository to a compute node running the hypervisor can take time in order of minutes. In addition to it, booting an image file can be a time consuming process if several applications are pre-installed. Use of caching to pre-fetch items that may be requested in future is known to reduce service latency. In order to overcome the delays in transfer and booting time, we prepare a VM a priori, and save it in standby state in a “cache” space collocated with the compute nodes. On receiving a matching request, the VM from cache is instantly served to the user, thereby reducing service time. In this paper, we compare multiple approaches for preprovisioning and evaluate their benefits. Based on usage data collected from an enterprise cloud, and through simulation, we show that a reduction of 60% in service time is achievable.