Reseaerch On Traffic-aware Virtual Machine Migration Algorithm

Virtualization, packaging applications into virtual machines, allows administrators to migrate or reshuffle virtual machines to accommodate workload spikes and resource shortages. As the network cost within virtual machine migration is alwalys overlooked, in this paper, we build a network-aware virtual machine migration system, with cloud simulator to simulate the cloud server and googledatatrace as end-user requests.Considering the impact of migration data and bandwidth on migration cost, we model migration cost and build a network cost evaluation model integrating communication cost and migration cost, while modeling migration-cost. Based on this, design and implement two kinds of network-aware migration algorithm with local greedy strategy, regardless of the server-side resource constraints and the upper application dependencies and underlying server topology, to optimize communication cost and minimize migration cost.Specifically, m Bef LM optimize net-benefits as the subtraction between communication income and migration cost, taking into account of the migration cost and communication cost in destination server select, to choose a virtual machine to migrate with maximum net-benefits. And m BRates LM optimize net benefit-rates as communication income of one unit migration cost, considering communication cost in destination server select, to choose a virtual machine to migrate with maximum benefit-rates.Overall, the design algorithms introduce network performance benefit and reference pre-migration network communication cost, considering the network performance benefit and migration cost brought by one migration action. Further, to improve the network optimization performance of algorithms, we set server resource thresholds to reserve resources for the subsequent arrival task, while developing a timer migration strategy to deal with the server resources jitter caused by user task resource demand changes shortly.By comprehensive theorety and experimental data, we test the performance of our network-aware algorithms on reduction of overall network-cost and task execution time, with existing algorithms App Aware, sandpiper and LVF, showing the network-aware algorithms can significantly optimize overall network cost under the premise of ensure total task execution time.In a word, we build a network-aware virtual machine migration systems, simulating end-user requests and the cloud server process, to provide a effective algorithms evaluation platform with real user task requests and realistic cloud processing. What’s more, the network-aware migration algorithms, introducing the concept of the migration cost and performance benefits, greatly reduce the whole network cost of data center.