This article can also be found in the Premium Editorial Download "Storage magazine: Comparing EMC Symmetrix DMX-3 vs. Hitachi Data Systems USP1100."
Download it now to read this article plus other related content.
The truth isn't sexy, but it sells
Vendors need to deep-six their marketing hype about virtualization and concentrate on the problems it can solve.
I've heard a lot of whining from various industry folks who blame users for not "embracing virtualization," i.e., parting with their money for some vendor's marketing pitch.
Instead of blaming users, vendors should look in the mirror. Storage virtualization technologies have been purchased and implemented successfully for years. The rest of the IT infrastructure will try to catch up (think VMware) and, ultimately, the only thing not virtualized within the data center will be the last guy standing.
Volume Manager was directly (and indirectly) responsible for billions of dollars of revenue for Veritas (and others). But nobody called it virtualization, although it's a perfect illustration of the technology. It makes something physical look like something else, and creates an abstraction between that physical thing and the stuff talking to it. If that ain't virtualization, I don't know what is. Veritas made Volume Manager the most widely adopted storage virtualization software by telling people what it did, not by tagging it with an ethereal marketing term. It also solved a capital cost-reduction problem; it let folks use the 4GB disk drives they bought because without it the file system would only address 2GB. That made those new "2GB" disk drives
Veritas engineers might have known they were virtualizing a physical disk, but their marketing and sales guys knew real people needed to make their new disks 35% cheaper than last year's, not 70% more expensive. Solving the capital cost issue is the first step in the "virtualization" continuum.
RAID is an example of the second step, increasing operating efficiencies. The Volume Manager solution let people use all the space on their disks, but data growth meant that even with spindle size doubling, they still had to add more units (of everything) to keep pace. Combine this with managing and maintaining 1 million instances of Volume Manager, and you see how a new problem emerged. RAID was intended to cut capital costs when minicomputers and mainframes used expensive $100/MB storage. The RAID guys--the "I" originally meant "inexpensive" but quickly became "independent"--built one big disk drive out of a lot of cheaper smaller disk drives, added parity in case one disk blew up and tried to solve the wrong problem, cost. When the cost disparity went away, folks in the RAID business needed to find another problem. RAID let you take 10 or 100 individually managed drives, shove them all in one box and manage one thing that looked like 100 "virtual" drives. The RAID business turned out to be pretty good once the industry realized it was solving a real problem.
VMware is an example of phases one and two. Running a lot of virtual machines on one underutilized physical machine means you don't need to buy a lot of physical machines. Managing 1 million virtual machines won't be any easier in the long run than managing 1 million physical ones, however, so that will be the next problem to be solved. The final phase is when we automate the mundane tactical tasks we perform on the piece(s) of infrastructure below. You don't stripe the disks in the RAID box anymore because the machines are better at that than you are.
Infrastructure virtualization is necessary and inevitable. But if you want to sell something, stop yelling at users and start understanding the problems they need solved. It's OK to mention that the higher up in the infrastructure the abstraction occurs, the more overall strategic value a customer will gain, but that has to be a secondary benefit, not the primary focus. You don't paint the house when it's on fire.
This was first published in January 2007