Dear IBM, EMC, and Hitachi: Do Real RAID 1

Dear IBM, EMC, and Hitachi,

Your storage virtualization devices are pretty neat. They let me buy lesser disk arrays and treat them as a pile of disks. They let me have different classes of storage and shuffle things around without the hosts ever knowing. They let me add cache to my storage architecture to make things even faster, and do intelligent caching in front of the arrays. And they let me do remote mirroring for storage arrays that wouldn’t do it otherwise. Sure, some of you use weird, proprietary, hard to maintain approaches like embedding the virtualization in the SAN, but that’s okay.

From time to time I need to do maintenance on my storage arrays. Because I use EMC disk arrays I end up always having to take a complete outage on the array, but all arrays have maintenance that is best done without I/O load. While your virtualization devices maintain a remote mirror, I still have to take an outage to cut over to the copy. I usually choose just to take an outage for a little while and get the work done. If you’d just make your devices capable of actively using a mirror my servers wouldn’t know the difference. Just like a RAID controller in a server, I’d like to pull the plug on an array and have the other array pick up the slack. You could even do neat RAID 1 things like spreading the read I/O out between the active copies, or keep track of changes during an outage and intelligently resync the copies afterwards. Call it SAN RAID or something. While you’re at it go all out and let people do SAN RAID 5. I bet there’s some crazy out there with a ton of data that would love to do that.

Wouldn’t that be cool? I bet you could sell some more devices that way. I’d buy a few.

Previous Post:

Next Post: