Prediction-based Optimization Of Virtual Machine Through WAN

Virtual machine migration is a key technology of cloud computing that required for cloud collaboration. Classic strategy of virtual machine migration, such as pre-copy strategy and post-copy strategy, can work well in LAN, but can not get satisfactory effect in WAN due to low and unstable bandwidth. In this paper, we propose a prediction-based strategy combined with Markov prediction model. The strategy we proposed optimizes the down time during virtual machine migration, and the strategy is discussed from snapshot level and virtual block level. In snapshot based level,snapshot increasing curve has been utilized to predict when to get the minimum snapshot. Then we can finish the iterative migration at that time to minimize the down time during migration. In block based level, a updated time series of each virtual block can be predicted, which helps to filter the set of blocks must be transferred. On the base of the strategy, a virtual machine migration system has been designed and implemented. The actual case and experimental results show that the strategy we proposed can achieve better effect compared with classic strategies.