In the server virtualization world there are three types of migration. The first, GuestMigration is the transport of a VM from one physical host to another. The most common example is VMware's VMotion. In this case there is no movement of the VM's data. The machine image is on a shared storage device accessible to all the physical hosts. The second, StorageMigration is where the data store of a virtual machine is moved to another storage system all together. This can be managed by the virtualization hypervisor or by the storage system software. It is typically used to migrate a VM's image to a less busy storage resource. The entire VM is moved at the point of initiation. The final type, Teleportation allows for VM images to be pre-seeded onto another storage device. Then when the time to move the VM occurs only the recently changed blocks need to be copied. This is ideal for migration across a wide-area network.

The shared storage resources that the virtual server environment counts on to house virtual server images can face many of the same challenges faced by server hosts. Storage systems can run out of capacity, storage processing power, or I/O bandwidth as well as experience availability issues. StorageMigration and Teleportation bring the flexibility of VM migration to storage by allowing rapid movement of the VM images themselves.

A RAID rebuild is an example of how storage availability impacts storage and virtual machine performance. When a drive fails in a RAID group the storage administrator has  to choose between dedicating most of the storage system’s performance to the rebuild process or to supporting on-going array operations. Focusing on rebuilding RAID groups can impact I/O performance of the virtual machines on that array. But making the rebuild a lower priority can expose all of these machines to potential catastrophic data loss if another drive fails before the slower rebuild cycle can complete.

StorageMigration may not help the situation since the entire VM needs to be copied from the failing and performance degraded array. Teleportation, as offered by companies like Nexenta in their NexentaStor NAS software, provides the ability to move a VM image to another storage platform, whether that storage is a separate system inside the data center or in another data center altogether, but the VM image can be pre-seeded. Then, once the storage is replicated a quick update is made of the changes that occurred during the initial replication. The VM can then be moved, with minimal downtime and without being impacted by the degraded array. Within a data center these moves can all be done quickly, depending on the network infrastructure.

In the RAID example above, critical, performance-focused systems can be shifted to a secondary array, so they can provide full storage performance to the applications they serve. While the VM does need to be shut down for the actual move, with the virtual machine image already in place on the secondary storage device, the updates can be sent and then that move takes only a few seconds. The virtual machine can be brought back online.

This Teleportation of a VM is more than just for the availability use case. It can also be used to balance I/O across servers or to simply populate a new system with VMs once it has been brought online. This allows storage managers to scale the virtual environment on commodity hardware matched to the specific use case, providing a “Google-like” loose clustering of off-the-shelf server components.

The most interesting use case is using Teleportation as a way to move VMs over distance. Storage Switzerland recently wrote an article entitled “Cloud Bursting with Distance VMotion” which discusses the high-end, real time movement of VMs. This requires high speed, expensive bandwidth connections and connectivity devices to make those bridges work. While ideal for some data centers, Teleportation over distance can make this a more widely used capability for all. Using the same process as described above data can be replicated over any distance, final synchronization to that data made and the VM teleported to the remote site. While once again, this does require that the VM be down during movement. That should be a fairly quick process and operations can resume soon after.

Cloud bursting is primarily a resource reallocation technology and teleportation can perform that reallocation function, but across a thinner WAN connection, as long as some downtime is acceptable. Due to the latency concerns with cloud bursting the distances between data centers has to be relatively short, less than a couple hundred miles. Typically this is not considered far enough to provide protection from a regional disaster. Teleportation does not have this distance limitation and can also be used as a disaster recovery aide. In this function it can be an alternative to the software products like VMware’s Site Recovery Manager by offering similar functionality integrated at the storage level and at significantly less cost.

Both remote and distance teleportation can be integrated into either the storage or virtualization management consoles. The entire process can then be controlled and monitored from the perspective of a single GUI.

Teleportation increases the cost savings derived in virtual environments, by leveraging storage software to do the bulk of the work, not the hypervisor or the storage hardware. This means that the storage hardware can be almost any off-the-shelf storage device that meets the organization’s reliability and performance needs. It allows for simple scaling by enabling VMs to be shifted from one storage platform to another as capacity and performance demands change.

George Crump, Senior Analyst

Nexenta is a client of Storage Switzerland