TY - JOUR TI - The design and implementation of cloud-scale live migration DO - https://doi.org/doi:10.7282/T3862DJJ PY - 2014 AB - Live migration, short for live virtual machine (VM) migration, enables a running virtual machine to move between two physical hosts without perceptible interruption in service. Live migration is an efficient tool for system administrators to perform system maintenance, load balancing, and fault management while allowing end-users to avoid costly service downtimes. Today, live migration between hosts connected by a local area network (LAN) has become a critical feature of enterprise class virtual infrastructure. We also expect live wide area network (WAN) migration, e.g., Cloud-scale live migration, to extend the scope of provisioning compute resources from a single data center to multiple geographically disparate data centers. Currently, Cloud-scale live migration is possible only through ad-hoc solutions using network file systems, proprietary storage array replication or software replicated block devices used in concert with more well known approaches for migrating memory. But this loose aggregation of mechanisms makes migration architectures complex, inflexible, and unreliable and performs poorly compared with live LAN migration in general. To overcome those deficiencies, we present a Cloud-scale live migration framework that integrates support for memory and storage migration over WAN and maintains much of the simplicity and reliability of live LAN migration. The main challenge for implementing Cloud-scale live migration is how to deal with the large VM data (especially VM storage data) transferring over WAN. To solve this problem, we propose a new “Migration over FedEx” solution to combine the benefits of both live LAN migration and transferring large amount of data via shipping portable storage devices containing the data. Our solution capped the total migration time into a bounded time period without increasing the downtime compared with traditional live migration. In the meanwhile, the total migration cost is greatly reduced especially for migrating large number of VMs. KW - Computer Science KW - Virtual computer systems--Maintenance and repair KW - Systems migration LA - eng ER -