Optimized delta compression in live migration of virtual machines

Abstract

Live migration becomes an essential feature for any popular hypervisor under consideration. But in many situations, because of huge amount of data transfer between servers, live migration is a challenging task at real time scenario. Ideally during live migration, the major performance matrices like down time, total migration time and amount of data transferred should be the minimum to keep all running applications in virtual machines uninterrupted. To achieve this, we analyze delta compression method to compress the VM snapshots while performing live migration between servers as a modification in KVM hypervisor. Delta compression drastically reduce the amount of data transferred and improve overall performance of live migration. Here we use LZ4 algorithm for delta compression instead of existing XBRLE. As per the experiments conducted for various workloads and input files LZ4 algorithm gives better results in terms of amount of data transferred, down time and total migration time and compression ratio as compared with XBRLE. In addition to this, to overcome the cache churning issue we suggest a four way set associative mapping for cache performance improvement.