|Institution:||Blekinge Institute of Technology|
|Keywords:||computer science - distributed computing; computer science - networks and communications; live migration; virtualization; load balancing|
|Full text PDF:||http://www.bth.se/fou/forskinfo.nsf/all/a5d9ea45dbf8369bc1257daf007951d3?OpenDocument|
We present two strategies to balance the load in a system with multiple virtual machines (VMs) through automated live migration. When the push strategy is used, overloaded hosts try to migrate workload to less loaded nodes. On the other hand, when the pull strategy is employed, the light-loaded hosts take the initiative to offload overloaded nodes. The performance of the proposed strategies was evaluated through simulations. We have discovered that the strategies complement each other, in the sense that each strategy comes out as “best” under different types of workload. For example, the pull strategy is able to quickly re-distribute the load of the system when the load is in the range low-to-medium, while the push strategy is faster when the load is medium-to-high. Our evaluation shows that when adding or removing a large number of virtual machines in the system, the “best” strategy can re-balance the system in 4–15 minutes.