Factor 3: Storage implications
The biggest single concern when moving to a virtualisation infrastructure involves storage.
While organisations may be able to get away with continuing to use direct attached storage for small development and test deployments, the larger a virtualised production environment grows, the less sustainable the approach becomes.
This situation is particularly true for those companies wanting to deploy tools such as VMotion, which enables the migration of workloads between different hosts for high-availability purposes.
Instead, a move to shared storage may be required, with storage-attached network (SAN) technology being the most popular in this context. Such technology is expensive but important, because virtual machines (VMs) are stored as disk images in the SAN.
In addition, every physical server on the network must be able to see each VM disk image to know when and where to allocate spare processing capacity should a problem occur with a given host, or should it need to be taken down for maintenance purposes.
Paul Mew, technical director of IT services provider Ramsac, says if an organisation consolidates servers down from, say, 10 physical boxes to two, each running five VMs, and then experiences a hardware failure, they may find themselves "in an increased risk position".
"The issue is that, because you now have your eggs in two baskets rather than 10, you'll have to try and recover half of your infrastructure all at once, rather than just a single server," he says.
Another problem is that if an IT department has begun virtualising its environment but has not factored in the cost of moving to or upgrading shared storage, then it can be a tricky to subsequently go back to the finance director and ask for more money due to lack of foresight.
But when purchasing a SAN, it is also important to check virtualisation software providers' hardware compatibility lists. Not all components of storage vendors' equipment are accredited to work in a virtualised environment, which may cause problems later.
It is just as important to think about sizing the storage infrastructure in relation to performance load. After virtualising their servers, some companies find applications run more slowly. This loss of speed is because the physical disks in a SAN can only process a given number of I/O requests per second, but VMs tend to generate so many that the disks cannot always keep up.
Workload analysis and planning tools, such as Novell's Platespin, can be used to get around this problem by creating a usage profile of how existing physical servers use memory, disks, processors and network bandwidth, and evaluating what capacity is likely to be required in a virtualised environment.
But such planning is also crucial in terms of storage volumes — or areas in the SAN in which data is stored. If multiple VMs running heavy workloads are all attempting to access the same volume, performance will inevitably be affected. Therefore, it is necessary that each volume is accessed by a mix of VMs with heavy and light workloads to ensure a balance.