Getting [Virtual] Religion.

Mark Ferelli

Hal Glatzer

MARK: In the mass-storage universe, nowadays, there are two magic words to conjure with: storage networking and storage virtualization.

HAL: Where's the magic in storage networking? I met the InfiniBand people at PC Expo, and they're staffing out by networking servers.

MARK: Yes, but they'll be in the storage universe soon enough. After all, what is a SAN if not a network?

HAL: True. But I think InfiniBand faces an uphill fight if it's positioned to displace SCSI and Fibre Channel any time soon. Whatever technical advantages it offers may be offset by the need to convince customers that they have to look at yet another alternative. Confusion isn't a good market strategy.

MARK: But it is a fairly common market condition! Look at that other magic word: storage virtualization.

HAL: Ahh, a place where confusion certainly reigns. I had a conversation with Michael Spotts, who holds the title of--if you can believe this--"Virtual Evangelist" at StorageTek. Of course, my first question to him was, "If you're the 'virutal' evangelist, who's the real one?"

MARX: Evangelist, huh? Did he make you believe?

HAL: I felt like reaching out and touching my disk drive and feeling the vibrations in the--

MARX: Hallelujah, brother!

HAL: But I'm still not converted.

MARK: Okay. Let's run through the catechism step by step. You already know that storage management is pretty much the same as information management today.

HAL: Yes. And I've heard you say that "data inflation" is turning information management into a large-scale, global-scale problem. I accept that too.

MARK: All right. So virtualization is a way to manage large-scale storage installations. When you virtualize, you're hiding the details of the individual storage subsystems from the user, by building a layer of abstraction between the operating system and the storage devices. I think of it as "conning" the OS into seeing something that isn't there.

HAL: Let me play Devil's Advocate.

MARK: I always thought you'd fallen from grace.

HAL: Michael Spotts told me--quoting him, here: "Not having to buy double or triple your storage capacity needs, to meet functionality, is where virtualization pays off." But don't you think that users are better-advised to get more capacity than they need anyway?

MARK: What users care about, in terms of capacity, are their application requirements and their growth potential. What they don't want to be burdened with is individual disk size, or the number of physical disks they need to get the job done.

HAL: But somebody has to know that and make the appropriate purchase decision.

MARK: The system administrator or the IT manager does. Definitely. They have to determine the physical aspects of the storage environment, including all the details of seek-times and cache size, and so on. But the users are more interested in throughput, responsiveness, recovery time, ...the applications aspects of storage.

HAL: Are you saying that that's where virtualization meets reality?

MARK: Just so. Virtualization draws a veil over the physical aspects. The only question remaining--and frankly, there are too many answers in the marketplace, right now--is where to hang that veil.

HAL: But why hide any of it? Why not just restrict access to some portions of the storage environment and open access to others? Keep the backup files away from anyone who isn't authorized to open or modify them, and give everybody else all the space they need.

MARX: That's the "old-time religion!" The object of virtualization is to create a homogeneous storage pool that looks the same to everybody.

HAL: But what for? The "old-time religion" is well established. Practically every PC user with a hard disk has it divided into logical drives. Isn't that "virtualization" too?

MARK: In consultant's terms, that's more "resource allocation" than "new layer creation." Look at RAID technology, where you have striping going on: You could call that "virtualization" too, by your definition. But it's still about allocation, rather than about conning the OS into seeing a pool of storage rather than individual devices.

HAL: Spotts spoke to that as well. He told me, "The efficiency in virtualization comes from having 'disk' look like 'tape.' In a RAID array, the OS writes to what it 'thinks' is tape, but is really disk; and it will transfer that data to tape later, under the covers, as it were. The RAID is 'lying' to the OS, because the 'tape' is really a disk buffer. And here's the extra benefit: You're writing and reading immediately. Data that isn't read immediately is rarely accessed again. The disk cache buffer is so big that if data is going to be read, it's probably still in the disk buffer."

MARK: Sounds good to me.

HAL: But it's awfully complicated. It consumes considerable storage resources to accomplish that trick. And for what? I still don't see where the "efficiency" comes in. Adding tape capacity is cheaper per gigabyte than adding RAID capacity. Where's the advantage of virtualiztion if it costs more?

MARK: There are several ways to implement virtualization that address that concern. In some cases, virtualization works in-band, meaning volume management within the storage devices themselves. In other cases, you have out-of-band solutions, where there are dedicated devices to handle the virtualization. That approach needs to be combined with a file system that's shared among all of the storage subsystems. Think "scaling" rather than "virtual" and you'll be closer to the concept. Some companies are now putting together specialpurpose network appliances that can act as virtualization platforms.

HAL: With what you and Mr. Spotts are saying, we've given our readers quite a lot to think about. I bet we'll get questions about this. Write to me at [e-mail] if you're still confused.

MARK: And write to me, [e-mail], if you've already implemented virtualization; we'd like to hear about your real-world experience.