DAFS Collaborative: Intel and NetApp Cooperate.

Editor's note: CTR Associate Editor Christine Chudnow conducted an interview with two charming industry evangelists and Direct Access File System (DAFS) co-chairs: Dave Dale of Network Appliance and Mitch Shults of Intel.

DD: My name is Dave Dale, I'm an industry evangelist working for Network Appliance and I'm focusing specifically on DAFS, which I also co-chair.

MS: I'm Mitch Shults: part-time industry evangelist, full-time director of business development for the fabric components division of Intel. My job is to make InfiniBand architecture happen. I do that along with a whole bunch of other people, I don't shoulder that burden by myself. I direct the investment strategy for our start-up equity investments in companies that are interested in doing things complementary to InfiniBand architecture. Now the direct access file system work that we view as directly complementary to InfiniBand architecture is actually independent, technically, of InfiniBand, per se. It's transport independent.

CC: Tell me about the DAFS collaborative.

DD: The DAFS collaborative was formed in June of 2000 and the sponsoring members are Intel and Network Appliance. The first real meeting was August 25 with 52 companies. And we had the first meeting of the specifications and to figure out what else needs to happen. The kind of timeline for the collaborative that we're shooting for is we would really like to have the specifications revision 1.0 complete and ready for submission for the summer of 2001. We haven't yet selected which is the most appropriate standards body, but I expect to have made that determination the beginning of the calendar year 2001.

CC: So you're not attempting to form your own standards body, per se. You do want to go through an existing one?

DD: Oh yeah.

MS: We observe that the industry has enough standards bodies.

CC: (Laughs) Isn't that true. And they can't agree on one darn thing, but that's a different story. What led Intel and NetApp to come together on this?

MS: I can speak from Intel's perspective and then Dave can cover NetApps' perspective. We were basically the inventors of the VI architecture which started actually in the mid-'80s if you want to go all the way back to its genesis with our parallel supercomputer efforts when we were in the supercomputer business.

CC: From the website the architecture says it was originally developed by Compaq, Microsoft, and Intel.

MS: They are the three co-authors. If the truth be known, a majority of the core technology came out of Intel server architectural labs. There were contributions from those companies and they might disagree with my assessment; you'd have to talk to them about the specifics.

CC: I'm sure I know what Bill would say. (Laughs)

MS: Bill couldn't come. (More laughter) We are interested, and always have been, in applications that fully utilize the performance enhancements that VI architecture provides. Our initial work with VI architecture was with database companies, for example, IBM, DB2, and Oracle who already have the ability to scale their applications by communicating between nodes very, very efficiently. Any time you reduce the latency of that communication, which is something that VI architecture does, you improve performance and its capability. But there's a far broader world out there of more conventional Internet applications that move files around or portions of files around that historically have been limited in terms of performance capability by the overhead of the protocol stacks that are being used on servers within the Internet data center to perform those information transfers.

NetApp came to Intel with the idea of creating and defining a new variant of a file sharing mechanism that was originally known as the I-file system. So the idea was to create a file system on the wire protocol that would allow very efficient communication enabled by VI architecture to be used for general-purpose file sharing with Internet/Intranet data centers. We thought that was a way cool idea and decided to lend our name and our technical assistance to the definition of that effort. With the understanding--and this was key to our agreeing to go down this path--that the results of our collaboration would be open and available to anyone in the industry, including Network Appliance's competitors and including Intel's competitors for that matter, to build interoperable direct access file systems filers and clients.

CC: Why?

MS: Our experience is that if you don't have that you don't get the benefits of open competition and broad acceptance that we're all seeking here.

CC: And of course Intel is working with a great many companies and needs to continue to do so.

MS: True. But Net App was the first to show up with this kind of proposal and we think it's an intriguing one. If someone else were to show up with something different, we'd consider it, but we'd encourage them to become part of this effort because frankly it's not in the industry's interest to have multiple competing things that do the same stuff. As long as NetApp and Intel are sincere as we are and our goal of making this open and interoperable, anyone should feel comfortable participating in this, including NetApp's competitors and Intel's competitors. Now that doesn't mean that anybody is planning to give anything away here. I'm sure NetApp intends, and I'll let Dave take over from here, to create very competitive implementations of DAFS. But they will be interoperable with other companies' implementations because users demand it. That's what has made NFS so successful and we're trying to replicate that success within the context of a high performance Internet data center enabled by VI architecture. Dave ?

DD: That was pretty succinctly put. I think NetApp's been tracking VI for a while and the next generation open cluster interconnect is really interesting, because from our perspective it enables us to combine all of the performance advantages of direct connect storage with all of the flexibility and data sharing and maintainability advantages of file access protocol where we have been specializing. So that in next generation data centers you really can't scale application computing independently from your data storage. You want to enable data transfers from applications while writing storage without OS intervention. We moved all of the overhead that's inherent in today's network file systems. It's possible in a data center because it's kind of a locked-down environment. We see VI as the enabling interconnect for the next generation data centers and we see DAFS as kind of the killer app for VI.

CC: Oh, thank you. I was going to ask you to clarify that a bit.

MS: That's the perfect sound bite.

CC: That's very good, I like it. Now those applications that will be built on top of or with DAFS itself, describe how those applications are going to benefit from this.

DD: Let me take a real wide application. An Internet Service Provider who has millions of mail accounts running over a whole bunch of application servers just running the mail application. What happens when that mail application is DAFS-enabled is that you get no fragmentation or reassembly or realignment, and data is moved by the application into the storage. The end result is that you're actually giving CPU cycles back to the applications because you adapt to the protocol between the application and the storage systems.

MS: Today when you're running, tearing through all these protocol stacks, sharing files, you're consuming a lot of the capacity of your system and what are really extraneous paths that aren't necessary. DAFS allows the full optimization. But I want to emphasize that the DAFS is specifically by design restricted to the confines of the data center itself. When you go beyond the boundaries of the data center into the local and wide area network, you are necessarily back in the world of TCP/IP and more complex protocol because you can no longer rely on the transport itself, the physical media, to have the characteristics that VI architecture requires, namely reliable, in-order delivery.

If you don't have that, then you need TCP/IP and you're back in the proven existing world of file sharing where NetApp is, I think, the 40 or 50 percent market signature leader. Dave?

DD: Yeah, something like 60 percent.

MS: So they have some experience in that sport. But they see the opportunity to take advantage of the growing existence of Internet data centers that has attributes that allow for more optimization than the broad Internet does for this type of technology.

CC: Before we sign off, is there anything either of you would like to add?

MS: Usually you guys ask about schedules and we usually dance about schedules.

CC: Yes, I know. I rarely ask for that reason. Do you want to tell me about schedules?

MS: Suffice it to say that there's no reason for anybody to hold back on this. There's a crying need for it from the customers. As soon as we're able to do it on an interoperable basis we'll get it done.

To learn more about Direct Access File System, look for an introductory article 9n DAFS in the October issue of Storage Inc.