Keywords: virtual machines, migration, storage, grid, cloud, wide area network.

Instead, they can lease less expensive resources from cloud providers. Therefore, IT services are increasingly spread across several data centers and sometimes it can be useful to migrate them. Unfortunately, VM migration techniques provided with virtualization environments are mainly designed for a LAN and therefore are not optimized to perform well with low bandwidth and high latency networks (i.e. WAN).

A key challenge in managing the migration of VM is what to do about resources that are associated with the physical machine that they are migrating away from. Indeed while memory can be copied directly to the new physical host, network connections and the access to the virtual disk demand more consideration. While the network connectivity issue can be solved by using a virtual network [2], the virtual disk remains difficult to manage. In fact, virtualization environments assume that the virtual disk of the VM is accessible from the physical machine to which the VM migrates. If a shared storage space is not used, the virtual disk will have to move along with the VM without interrupting the latter.

In this paper, we reviewed several storage solutions that enable the migration of VMs by taking care of the continuous access to the virtual disk. The solutions based on distributed file systems (DFS) seem to be the most incline to provide WAN-ready migration systems. Indeed, they are designed to manage thousands of nodes and they are highly flexible. In this way, virtual disks are stored as a form of a file within the shared storage space provided with the DFS, and the latter ensures the continuous access to them even while the VM migrates from one host to another. Unfortunately when the network latency increases, the performance of I/O operations through virtual disks may decrease as well. We ran some tests with the benchmarks MAB and IOzone in order to evaluate the efficiency of the access to virtual disks through the network while the latency increases. We have shown that a DFS like ViSaGe or Lustre supports WAN migration with a good efficiency, but as expected the performance decreases as soon as the latency goes up. The main issue is that if there are a lot of I/O operations to the virtual disk, the performance of the VM may decrease if the latency with the virtual disk is too high. It is possible but not recommended to use the replication mechanisms provided with DFS to make a copy of the virtual disk closer to the migrated VM. But replication has a cost to ensure coherency of all replicas, that is why we believe that a truly migration of the virtual disk is required with DFS to provide better flexibility and more efficiency to the migration of virtual machines on a WAN.

1

P. Liu, Z. Yang, X. Song, Y. Zhou, H. Chen, B. Zang, "Heterogeneous live migration of virtual machines", International Workshop on Virtualization Technology, 2008.

2

T. Wood, P. Shenoy, K. K. Ramakrishnan, J. Van Der Merwe, "CloudNet: A Platform for Optimized WAN Migration of Virtual Machines", University of Massachusetts, Technical Report 2010-002, 2010.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £85 +P&P)