SearchStorage: NetApp has taken a varied approach to Flash, where it's released solid-state disk (SSD) support as both an expansion of read cache, as well as a disk drive. I'm wondering how you see those two use cases playing out.
Dave Hitz: Flash is fascinating stuff. It's about 10 times more expensive than disk drives—a lot cheaper than memory, though. It's 10 times more expensive than disk drives, but for random-access stuff about 100 times faster. Awesome bang for the buck for performance.
The tricky thing about Flash, though, is that it's so expensive; it's not appropriate for all storage for most applications. In a lot of applications, if it's a relatively small amount of data, it can be. We've seen that with iPods switching to Flash, and more and more PCs are switching to Flash—I'm sure the next PC I buy or the next laptop will be all Flash. In the context of a large storage system, though, only a very high-performance, high cost database kind of app that really, really needs that performance would probably be worth paying that 10x cost for the whole amount of storage.
One thing I will say about Flash—everybody in the storage industry understands that this is a game changer. We're fighting a little bit about [it] -- some people think you should use it this way, some people think you should use it that way, but realistically, over the next five years, all storage vendors will be including Flash in their systems in various ways and trying to figure out what's the best way to do that.
SearchStorage: Some people predict a new tiering system where high-performance data goes on Flash and disk drives can be low rpm, big SATA capacity for persistent storage. Do you see that happening?
Hitz: Yeah, I'm one of those people. I wrote a blog [and] the title was 'Flash is the new disk and disk is the new tape,' and I did some back-of-the-envelope calculations. If you remember those big old mainframes that you used to see in science fiction movies that had the tapes on the front…relative to the CPU speed of those old mainframes, disks today are slower than those tapes were in terms of how much seeking was going on per cycle of CPU.
One of the biggest bottlenecks in computing is that we've got these super, super fast chips—think how many billions of operations can occur between the next seek that you do on the disk—it's just untenable. I absolutely believe you're going to see it. We already have that kind of conversion of tape backup going to disk anyway, so in a sense we were losing a tier. I think Flash is going to help us make back the tier that we were losing.
SearchStorage: Some people believe the SAS-2 spec will finally put SAS into prime time in terms of external storage systems and possibly overtake Fibre Channel (FC). What's your take on that?
Hitz: My take is 'I'll believe it when I see it.' I don't mean to be skeptical here, but it's very difficult to tell how disks end up being priced and how they end up being deployed until they actually get into volume. We've been burned by being the early guy to rely on a new technology – we'd rather add it to the portfolio and see how it plays out from a cost, performance and a reliability point of view. We support all these different disk drives, we see what's cheap from the vendor and the disks themselves are pretty much commoditized.
SearchStorage: NetApp's core company values are listed in an appendix at the back of your book, and the first one is 'Driving customers' success.' Do you still get to talk to customers in your current role? What's top of mind for them right now?
Hitz: Right at the moment, I have to say from a business perspective the [economic] downturn is top of everybody's mind. [Here are] two of the ways that NetApp tends to work with customers: one of them is to really help them figure out where they are trying to take their data center -- how they'd like to architect things, what's the big picture [and] future grand plans; and the other one is how do we really roll up our sleeves and make things more efficient? In this economy, I have to say that [the] second conversation is much more interesting to most people than the first. It's all about going pretty easy on the budget.
One of the biggest trends out there is server virtualization, and the architectural changes that drives in the data center. It turns out getting the storage architecture right is really critical behind that. That's one transformation I think will be accelerated by the downturn. I think the biggest barrier that people have to making that transfer is 'Is a virtual machine really as good as a physical machine? Is it going to be reliable enough?' But I think in a downturn like this there are three words everybody has in their mind: 'Good enough, considering.' Is VMware as good as a real system? I don't know, but it's good enough considering my budget is flat or down. I think you're going to see a lot of conversion just because it saves so much money.
SearchStorage: In the book, you describe NetApp's success at the rise of the Internet as catching a really big wave, and that the secret to success in business is identifying waves like that. What's the next big wave for NetApp?
Hitz: I see three driving trends that are going on right now in the industry. No. 1 is how do you architect data centers? And the trend behind that is the conversion to server virtualization and all of the other implications of that—how does that change the storage requirements, how does it change the networking requirements, that transformation. I think to around 2015 is when we're going to see that play out.
The second big trend is whether or not you should build a data center at all. And that's the whole question of outsourced computing, cloud computing, which I define broadly as you solve the same business problem you solved before, but somebody else buys all the equipment and you just rent it and access it over the Internet. The big stuff, we keep in our own data center, kind of like [when] people kept their mainframes as stuff migrated into Unix and Windows. But I think more and more stuff is migrating into a cloud type of environment; we're barely even noticing how much it's happening.
In terms of how storage systems are built, Flash is clearly the driving trend there.
SearchStorage: When it comes to the cloud, there are people in the market who remain skeptical and say Storage-as-a-Service (SaaS) and storage service providers (SSPs) didn't fly the last time we tried this, back before the Internet bubble burst. Why is it going to work this time? How would you answer that?
Hitz: My answer would actually be to share some of that skepticism. I think that one of the least interesting services to supply separately as a cloud is storage services and here's the reason. If you think about the end terminal, the bandwidth from there to the CPU is relatively low. We do that over WANs all the time. On the other hand, the bandwidth between the CPU running the app to the storage, that's extremely high. So it makes a lot of sense to access the CPU out over the Internet, but not so much sense to access the storage. I think you'll see much more cloud services, but the storage will still be in a data center next to the CPU. The one exception, I think, is for backup and archival kinds of data where you're sending the data periodically and it's not as high bandwidth. I think cloud services for that can make sense.
To download a podcast of our full interview with Hitz, with more on NetApp's evolution and the writing of his book, visit our Storage Soup blog at IT Knowledge Exchange.