Newsman's startup eyes multimedia nets

As the demand for delivering data across distributed network architectures increases, it's driving a new generation of storage technologies. Network Appliance's CEO Dan Warmenhoven and VP of Strategic Marketing Ray Villeneuve sat down to discuss the company's role in shaping the future of networked storage.

Villeneuve: We've been arguing for years that the distributed model is the right way to view a datacenter. But the issue is how are you going to think about data management? The network world figured this out a long time ago. You don't necessarily need people where the network equipment is. So they had to figure out how to centrally manage distributed data. The data management world has not gotten there, and that's really been a core component of our strategy: How do you manage distributed data?

Villeneuve: That's been a fundamental piece of the architecture since we brought the first system to market. The multivolume capability probably showed up around 1997. That's when they had the first real virtualization capability. But even before that, the notion was to take all your storage and think about it as one lump. That's the image we presented to the world. Dynamically expandable volumes were a fundamental piece of the architecture since our first product was released.

Villeneuve: We've taken that to the next level in our recent product releases -- everything from virtual filers where you have multiple virtual machines in one physical one through to this notion of what we're doing for the block services with virtual volumes or virtual disks. We even have a capability to look at a volume in this new implementation as a file, so you can back it up like a file. So virtualization is all about aggregation and segregation. It's just that simple. I want to be able to look at all these disks together as one resource pool. And then the second piece is I want to be able to take and create logical disks.

Villeneuve: The biggest [challenge] is it puts a lot of pressure on customer budgets. If customers spend money upgrading their servers for the higher performance needed to handle XML, then they don't have money left over to buy more storage. So, especially in tight economic times, we're fighting for a fixed pie. We spend a lot of time talking with customers about how to save money through storage consolidation. So we don't actually compete with XML head-on, but we co-exist with it.

Warmenhoven: That's right. Isn't that terrific? I love that.

Warmenhoven: Yes. When you get to enterprise caching you're largely talking about caching Microsoft content. The enterprise, especially at the edge, is Microsoft dominated. You might find Unix servers in all the datacenters, but at the office locations, where the caches go, the distributed environments are almost purely Windows. And that's just XML, but why don't you go ahead and cache Windows files?

Warmenhoven: Microsoft is an environment, right? And they dictate the weather and the climate and the mix of the atmosphere and whether there's oxygen or not. .Net is going to create another environment for distributed applications.

Warmenhoven: I had a real in-depth conversation with some of the guys at Microsoft recently, and I don't think they would say that. We found that we were using the same words to mean two different things. We were talking about our file system, we were talking about how we optimize utilization disks to control formats on disks and get very rapid access information from the disk. And they're talking about file systems as a set of services provided to an application.

Warmenhoven: We are working with a number of application vendors. For example, Accenture has just published a report that showed how we accelerate end-user response time in SQL environments. The unique thing about SQL 7 is the Java applets run at the edge. There's a zillion and one transactions occurring between the client system and the server system, and half of those are just trying to get the next Java applet out. Well, why don't you cache all the Java applets? What happens if you cache the frames? The only time you would have to go back to the server is for the variable data components. You can see how this might be applied to other environments.

Warmenhoven: Right. I'll serve it up and you'll get it at LAN speed as opposed to WAN speed. And guess what, a single server back there can't serve that many applets anyway because every server applet request looks like a transaction. It's just going to bog it down. However, we don't actually run the applet. All we're doing is serving the applet. You're still running it at the client side.

Warmenhoven: I think we've always had a different model of the storage world than anybody else. First of all, the issue is not storage. I used to argue that we are not a storage company, we're a data management company. We're solving data management problems. Customers ask for complete solutions. If you're going to manage my data, I want you to hold my data, I want you to support it. It's the one throat-to-choke kind of model. The original business plan for this company was not to ship hardware, it was in fact to be a pure software company.

Warmenhoven: I don't buy that it works. I understand the theory, but that sounds to me like just another service play. I haven't seen a lot of customers jump up and applaud. A fine business strategy for them, but I'm not sure it's going to get a very strong reception in the marketplace.

Warmenhoven: iSCSI [Internet Small Computer System Interface]. Our view is that iSCSI will overtake Fibre Channel, and that the network convergence for both NAS [network attached storage] and SAN [storage area network] will be on an Ethernet TCP/IP base, 10-Gigabit Ethernet. And that the style of access between the server and the storage subsystem will be determined by the application. So you've got a server with a NIC [network interface card], or an HBA [host bus adapter], depending on your preference. That's attached to an Ethernet network, a storage subsystem, and over that you can access file or block data. I think when a pure iSCSI world comes it will overtake Fibre Channel. It's a lot less expensive.

Warmenhoven: A product. Can you find one? We are attempting to ship real iSCSI, but it's not "real iSCSI." It's iSCSI-like, but I couldn't find an iSCSI NIC or HBA.

Warmenhoven: We would like to ship a native iSCSI that is industry standard-compliant, based on a smart NIC, a TCP offload engine.

Warmenhoven: We're not far away. It may be measured in an amount of months. The implementations are largely complete. We had one run in the lab for five minutes. It was an endurance test -- it fell over in five minutes. So things like error recovery aren't quite developed yet. I believe this is going to be the year, 2002, when most of the vendors in the industry announce iSCSI products. I think 2003 is going to be the year when most customers seriously start evaluating it as an alternative. I think in 2004 is when it really starts taking off.

Warmenhoven: No, I don't think so. I think there are going to remain three segments of vendors in this area. It's going to be real storage guys who ship disks, and we're in that category. Then there's going to be the networking and switching guys, and that's where I put Brocade, Cisco, and Rhapsody. And then I think there's going to be data management services, which will integrate this technology across the server environment. And I don't think you're going to see a lot of crossover between those three.