Performance is clearly an issue in backup, however, performance in general is a key storage array feature for virtualization.
Storage arrays have been developed to process large volumes of I/O, which can be either sequential or random. In virtual environments the I/O is typically random and this doesnt work well with DAS storage, which would require more expensive, high-speed drives.
Storage arrays can benefit from large numbers of disks (as it is a shared environment), dedicated cache and multiple I/O connectivity, all of which both improves performance and delivers a more consistent I/O response time.
As more workload is virtualized, the hypervisor itself becomes a significant part of the support effort, because it is the platform that is tied to the hardware itself.
Boot from SAN enables the hypervisor to be disconnected from the hardware and allows a single hypervisor instance to be booted on any server; it also allows the hypervisor to be replaced with another instance that could be (for example) an upgraded version.
By removing the boot device from the server and placing it on the SAN, the server holds no state information and so becomes a commodity. This is most easily demonstrated with the use of blade servers, where multiple physical servers exist in a single chassis; they can be added or removed from the blade infrastructure at any time. Ultimately, blade flexibility is served best with shared SAN storage.
Both hypervisor and boot disk stored on SAN; hypervisor can be booted from any physical server.
VAAI defines a set of API calls that are implemented within the storage array through amendments to the SCSI protocol. Most notably of these are the following:
Block Zero implemented as Write Same in SCSI, this pushes the task of zeroing out large blocks of data down to the array. In fact, Write Same could be used to write any values over a large range of data, however its most useful to vSphere to write zeroed out data when creating new virtual disks (VMDKs).
Full Copy implemented within the array as SCSI EXTENDED COPY, this feature allows bulk movement of data both within and between storage arrays, taking the load off the vSphere hypervisor when performing storage vMotion or guest cloning functions.
Hardware Assisted Locking (HAL) this moves the SCSI hardware lock from the LUN to the block level, improving performance on certain vSphere operations that require locking for data integrity. However, HAL will potentially resolve the issue of LUN replication, allowing I/O on both sides of a replicated LUN pair.
Example 1: EMC VPLEX
EMCs VPLEX product virtualizes the storage LUN and permits I/O to either side of a replicated LUN pair. This enables virtual guests to be moved between storage arrays (typically in geographically distant locations) with no outage and without waiting for data to be replicated.
Example 2: Compellent Live Volume
Compellents newly announced Live Volume feature enables a single logical LUN to be spread across multiple storage arrays. The LUN can be associated with one array and dynamically moved to another in order to meet workload balancing or DR requirements.