HPCwire
 The global publication of record for High Performance Computing - LIVEwire Edition / November 20, 2003: Vol. 10, No. 3

  |  Table of Contents  |  

Features:

INTERVIEW WITH TODD MATTERS, INFINICON
By Tim Curns, Assistant Editor, HPCwire

HPCwire: What new developments are taking place with InfiniCon, most notably here at SC2003?

Todd Matters: Well, as you know, earlier this summer we introduced the InfinIO 3000 Switch Series, which was a monumental step forward in terms of providing 10gb fabric for HPC solutions. It was a family of products that really brought very dense packaging and really made InfiniBand practical in the HPC environment. We have had a lot of end user success with that product, and anticipate the same with our next product, the 9000 Series, which we've announced this week. We see the 9000 as bringing InfiniBand into mainstream HPC and commercial sectors. Basically the 9000 features new silicon technology from Mellanox that we are leveraging with our system architecture design to continue to drive the price point down to about $700 per port list price. That's about a 30% reduction in list price currently. We see that price point really removing a lot of barriers of deployment and providing an impetus for InfiniBand into more of the mainstream. We're already seeing tremendous performance and scalability with the current products.

HPCwire: What areas or sectors do your products affect the most?

Todd Matters: It's a combination -- a lot of universities, and research institutions. They've kind of been the early adopters. We're beginning to see a lot of uptake in the commercial market as well, we see the commercial market really blossoming in 04. There are a number of market drivers that are allowing the types of switch products that InfiniCon develops to really flourish in the HPC environment. We're seeing the continued commoditization of computing resources, so today it's economically practical and feasible for a very large number of institutions to deploy64-120 node fabrics. Only five years ago, the cost of the compute resources was too exorbitant. If you combine the low cost of compute resources with the maturation of distributed applications the scalability improves dramatically. So large clusters are affordable, and the applications are supporting that wide scale out. So where we come in, is the interconnect. You have to tie all that stuff together, you need the right bandwidth, and you need the right latency. Those three factors together is allowing us to put a product into the HPC market and driving it into the mainstream.

We haven't announced details, but in that family of switches (9000) will be a very dense packaging that would provide well over one hundred 10-gigabit ports into a single chassis. We see that being very instrumental, again as these fabrics really continue to grow, proving the scalability. We see that product also working to drive InfiniBand into the mainstream HPC and commercial sectors.

HPCwire: Can you please tell us a bit about yourself, perhaps offering a brief chronology of your achievements? I do know that you co-established InfiniCon.

Todd Matters: I've been in the industry for over 20 years. I went to work in the early 80s developing SNA products. I completely fell in love with networking and high performance networking. One of the career changing things that happened to me was I went to the Federal Reserve Bank. I worked on a product for about two years, a very large, enterprise networking product. You know it worked, and it was in the lab, and I went to the Federal Reserve Bank and I was just in awe: they were using my product to do billion dollar fund transfers. Going in and seeing the fruits of your labor for over two years .that was a career changing experience for me.

I continued to become sensitive of end-user and customer sensitive environments. My partner, Phil Murphy, and I had been tracking developments in I/O and networking. We were predicting that for standard high volume servers to really gain the kind of penetration into the enterprise market, they would need a new I/O networking system. We saw a need for a fundamental change in the way I/O networking was done. We started InfiniCon specifically to address those needs.

HPCwire: What other fundamental changes needed to be made in the near future?

Todd Matters: Most of my guys were in enterprise, but about 18 months ago we started getting into HPC because it ended up being a very good fit for the types of products that we were building. Being in the HPC environment now for the last 18 months, and seeing the fruits of our labor is immensely satisfying. We see us continuing to focus on the HPC environment.

HPCwire: What's in the works for InfiniCon? One aspect we're really focusing on, its almost a cliche now, is fabric management. I see fabric management being absolutely essential to the scale of these HPC environments. A major effort of ours right now is fabric management -- fabric visualization, fabric reconfiguration, operation of the fabric, diagnostics, and those sorts of things. We've have some really exciting quarter or so. There's a whole bunch that we can't talk about yet that we're really excited about. We'll hopefully be unveiling this stuff in the next quarter. This is such an exciting time for us. Seeing customers excited about our products is immensely exciting for all of us as well.


Top of Page

  |  Table of Contents  |