Whats Appliance-Based and Switch-Based Virtualization?

Q: Whats the difference between appliance-based and switch-based virtualization?
A: All storage virtualization is based on specialized software. The difference between appliance-based and switch-based virtualization is really just a question of where vendors choose to put the software. This choice is usually based on how they want to position themselves in the market. The original virtualization players like DataCore and FalconStor were pure software vendors, so they chose the appliance route. They put their software on a box that is just a regular Intel- or AMD-based server running an OS like Linux or Windows Server. On one side network cards connect the appliance to the host, on the other side they connect it to the SAN array. IBMs SAN Volume Controller works the same way except that it comes already integrated on IBM server hardware. You can also put this functionality back in the array itself. For example, in a Compellent array the array controller is just an off-the-shelf server running virtualization software. Switch-based virtualization on the other hand puts the software on a control processor that resides within a SAN switch. The software actually functions as an out-of-band controller that tells the ports on the switch what to do. This is what EMCs Invista does. They basically take a Clariion storage controller that runs Windows Server and attach it to a virtualization-enabled Cisco or Brocade SAN switch to tell the ports what to do.

Q: Can customers really use storage virtualization to manage multi-vendor storage environments without sacrificing functionality?
A: Yes, running consolidated multi-vendor environments is quite feasible with storage virtualization. However, there is often a political problem in that the big array vendors like EMC, Hitachi or IBM just hate seeing their gear running behind someone elses virtualization platform. Of course theyre happy to virtualize the other guys stuff, but they will fight tooth and nail to persuade their own customers not to do it to them. In fact, this has always been a problem for the pure-play vendors like DataCore and FalconStor, which is why they have refocused their business away from storage virtualization as such to specific solutions implemented on top of it. For example, FalconStor is focusing on virtual tape and continuous data protection. DataCore is getting a lot of its customers from server virtualization.

Q: How does block-level storage virtualization work with server virtualization like VMware?
A: A lot of smaller VMware users with say just 10 or 12 server boxes running the ESX hypervisor would like to have the reliability of a SAN for all the virtual machines they create on these boxes. But when they look at the price of even a small SAN array they dont like it. So an alternative is to take a pair of the servers they have freed up thanks to VMware and repurpose them as de facto storage controllers. They can put storage virtualization software like DataCore on these servers and then hang some cheap JBODs ("just-a-bunch-of-disks") off the back. Presto, they now have a 20 or 30 terabyte redundant virtualized block-level storage array without paying for a SAN. This assumes they are running applications that need block-level data access like Microsoft SQL Server or Oracle. If their virtual machine applications only need file access they can buy a NAS or a just a plain old file server.