Where should storage virtualization be done -- Host, network, array?
Let me discard host-based deployments upfront. Host-based storage virtualization
was the first virtualization product in the open systems world. It came from Veritas Software Corp. more than a decade ago and was designed to allow a single server to "look at" a number of physical devices from different vendors in a consistent fashion. It was a single server solution, and it is not adequate anymore, so you should tackle storage virtualization in the network or in the controller.
The only example of controller-based storage virtualization is from Hitachi Data Systems (HDS). There are pros and cons. Virtualization is being done inside the storage system (e.g., an HDS TagmaStore), so a customer committed to a homogeneous HDS environment may see that as a great advantage. Just remember that all storage traffic, from an HDS or non-HDS box, has to flow through the virtualized storage array controller.
If you're highly heterogeneous, a network-based solution will probably be more suitable. Network-based storage virtualization can be accomplished through an appliance, based on an Intel server, an intelligent switch
or a purpose-built appliance (PBA). All three techniques have a place in the network, but your choice of approach depends on the storage vendor that you have a current partnership with and the level of network performance that you require.
Check out the entire Storage Virtualization FAQ guide.
This was first published in August 2007