Avoiding the Disk Bottleneck in Deduplicated VM Transfer

Abstract

This paper presents an optimization mechanism to increase the performance of cloud services that transfer groups of deduplicated virtual machine (VM) images. This is necessary as the naive data transfer approach for groups of deduplicated VM images is extremely inefficient as it generates highly random disk access pattern. The optimization mechanism presented significantly improves the VM image transfer performance through intelligent buffering, block-level transfer scheduling, and data indexing to increase disk access sequentiality.