Features:

INTERVIEW WITH CLIFF MILLER, MOUNTAIN VIEW DATA
By Tim Curns, Assistant Editor, HPCwire

HPCwire: What news does Mountain View Data have for SC2003?

CLIFF MILLER: Well, we're working with AMD, we are in their booth. We just did a press release this morning about Powercockpit being enabled to deploy 64-bit images to cluster environments on top of AMD processors. That's big news to us because with HPC, AMD has really taken off. I think the Linux community likes AMD a lot. That's our big news for this week.

Last week, we made an announcement about Powercockpit being used at the SCI Institute (Scientific Computing and Imaging Institute) located at the University of Utah. They've got about 70 world class researchers doing scientific visualization and scientific computation there. They have three clusters that are managed by Powercockpit. They've been using it for a year or year and a half and seem to be happy campers.

HPC: What other areas do you feel you need to focus on to continue your success?

CM: There are a few things that we've been working on. One is our technical staff is extremely good at low-level Linux stuff. Things that are close to the hardware, stuff that involves the kernel. We've got several developers who are as good as they come, and when you look at the products that manage clusters, they tend to be higher level, dealing at the application level. We can do that too, but we're going to focus on the lower-level stuff that will allow administrators to have very hands-on control of the hardware. This is something that no one else is really doing.

In addition, IPMI is a relatively new standard and many of the servers shipping next year will have IPMI chips from the tier one hardware manufacturers. If, for example, the OS crashes and you need to reboot your machine, you can bypass the OS and go directly to that IPMI module. That kind of functionality will become extremely important I think over the next year or two.

The other area that we're working on is the higher-level. In the last six months or so in North America, we've seen Linux really move into the corportate environment in the data center, so its not just being used for web serving and mail serving. It's being used for cluster databases, for example. So, instead of having a big Sun machine running Oracle, you may have a 32-node cluster running DB2 or Oracle. That business according to some of our partners has increased over 100 percent in the last six months. So that's an important area for us -- building modules for Powercockpit that allow administrator to manage clusters of machines running things like databases, application servers and so on. In general, that'll be the corporate environment, rather than HPC.

HPC: What prompted you to create MVD?

CM: I started MVD about three years ago. Essentially, having been in the Linux environment, and having built products based on Linux, I assembled a team of some top notch developers to work in a few new areas. Initially, we were working in network storage, and we still do that. About a year and a half ago, TurboLinux [Miller's previous company] didn't have the momentum it had in Asia. So it ended up closing operations in the U.S., selling off its Linux business to an SI in Japan, and putting up for auction Powercockpit which was developed by an excellent team of developers and scientists that I pulled out of LANL about 4 years ago. That group was in Santa Fe and we set up a lab for cluster developmenet. The upshot of that was Powercockpit as the resulting product of that. So this was a first rate group of people who had been doing this at LANL for quite some time. Powercockpit is a highly robust software. One of the best. It has never crashed in a user environment. There are very few commercial applications you can say that about, beta or production that never crashed. It's extremely robust, very solid. We have very few support issues. It's not "there's a bug," it's "we'd like to have this added functionality." So we work with our customers the hardest.

Anyway, about a year ago we ended up acquiring Powercockpit and continued development with that. What's happened is you have these very big supercomputers, these monolithic computers being replaced by clusters of tens, hundreds or thousands of Intel or AMD based PCs. A missing component here is software to manage it. With a monolithic supercomputer, if you're going to install software on it, you don't have a big problem. It's kind of like installing software on a PC. That may be an oversimplification, but not much. You install the software and away you go. Now if you have 1,000 nodes, you may have to take a floppy or CD around to a thousand computers. By the time you get to the last one, your number three may have broken down already. It may be too time consuming to manage these.

Our software helps to automate these tasks. If you go around SC2003, you'll see that probably 80 percent of these companies are hardware companies, networking or server companies, or what have you, but there are very few software companies. If you talk with people about what they're doing with their clusters, they'll tell you all the cool stuff, but if you dig deeper into what they're doing to install software, what they update things with, what they use to monitor the different nodes, what happens when one goes down, you'll find a lot of dissatisfaction in what's out there. So, there's a huge gap that needs to be filled. That's what we're trying to fill.