What are storage vendors doing right now to make their products more virtualization-friendly, and how important is VAAI storage to the future of virtualization?
What irritated me about VMware is that about two years ago, it came out with VAAI [vStorage APIs for Array Integration], a set of nine primitives or nine commands in SCSI, that it had written. They didn't get [prior] approval from the ANSI T10 technical committee, which oversees the SCSI control language used to talk to storage devices and other peripheral equipment. VMware just arbitrarily produced nine new commands, and the idea was, "We're going to reduce the workload that we're having such a hard time processing in VMware by offloading that to controllers of storage arrays that can do that job for us." So if you need mirroring, why have VMware administer the mirroring operation? Instead, offload that with a specialized instruction to an array that has on-array mirroring, and it will mirror to another physical array, sort of through the back door. That was supposed to alleviate about 80% of the congestion that they were experiencing inside the server virtualization environment. They were just going to offload all this heavy lifting down to the array controllers.
But the problem with that theory is what happens if you send that command and data to an array that doesn't speak VAAI, that doesn't have those primitives, and that isn't VAAI storage and aware of those changes made in the command and control language? And who gave [VMware] the right to start making changes however they see fit? That's the biggest hubris involved in the whole situation. The bottom line is that the industry responded fairly quickly, and there are now many VAAI storage options. The question is, was that the vendors' responsibility to make storage easier to do with virtualization, or was that just the vendors stepping up to fix a problem that was created by the hypervisor vendor in the first place?
This is part of my bigger rant about VMware. You go to VMworld and there are 190 booths with vendors who are part of the ecosystem for VMware. But if you sit down with any one of them to do an article, they'll tell you they're fixing a problem that VMware breaks. So it's a whole ecosystem based on the problems created by the primary vendor. That doesn't seem right to me.
Right now, what we're coming to is a very bad situation. We're seeing it in cloud technology, to a certain extent, where each vendor of a virtualization hypervisor has its own ideas about how storage should work underneath them. And now you have VMware talking about doing its own storage hypervisor as part of its kit, and that means I'm going to have to segregate a portion of the storage infrastructure I've spent the last 10 years aggregating into a common Fibre Channel fabric. So I'm now going to have to cut that up to serve those vendors that are serving VMware? And then who knows? I may have to cut it up some more to serve those servers that are running Hyper-V, those that are running Citrix, and those servers that are running Oracle. It's ridiculous.
This was first published in April 2013