In the realm of Rock Stars, there are One-hit Wonders, Divas, Boy Bands, American Idols, Crazy Hearts, and Legends. Our insideHPC Rock Stars are clearly an elite group of industry luminaries and thought leaders, but even among this group, few have attained the legendary status of this month’s insideHPC Rock Star.
The old timers in the community of course know Steve Wallach. As co-founder of Convex Computer Corporation, he was well respected throughout the computational science and the investment communities. His technical leadership was chronicled in Tracy Kidder’s Pulitzer Prize winning book, “The Soul of a New Machine.” Convex was eventually acquired by Hewlett-Packard, and Steve took on the role as Chief Technology Officer of HP’s Enterprise Systems Group.
Most recently, Wallach has once again drawn the spotlight as the Chief Scientist, Co-Founder, and Director of Convey Computer Corporation.
Wallach has 33 patents and is a member of the National Academy of Engineering, an IEEE Fellow, and was a founding member of the Presidential Information Technology Advisory Committee. He is the 2008 recipient of IEEE’s prestigious Seymour Cray Award.
It is with great pleasure that we present to you the newest Rock Star of HPC, Steve Wallach.
insideHPC: You have such a rich history in this community and have been involved in so many milestone activities – what would you call out as one or two of the high points of your career – some of the things of which you are most proud?
Steve Wallach: One high point in my career was starting Convex Computer, a company known for its “easy to use, affordable supercomputing” technology. The tagline when Convex started was: “A minicomputer version of a Cray, but program like a VAX.” At the time, very few people believed in the “program like a VAX” part. Today, the compiler technology that Convex developed, with the help of the late Ken Kennedy of Rice University, is considered standard.
The second high point is being able to give something back. I was a founding member of PITAC (Presidential Information Technology Advisory Committee). We helped to increase NSF budgets by hundreds of millions of dollars. Also, I’ve served on various government studies on high-performance computing. I spend lots of time in the greater DC area. One associate even went so far as to suggest that I get an apartment in DC, so it would be easier for me. I politely declined.
insideHPC: What are your thoughts on how we can attract the next generation of HPC professionals into the community – and provide them with the experience-based training that they will need to be successful.
Steve Wallach: That’s a tough one. Perhaps we should follow the lead of Apple’s App Store (or Android): Easily available and easy to use.
First of all, we need to increase the productivity of programmers. This generally means system manufactures need to employ a higher level of co-design (designing hardware and software together). I believe that, as part of all major RFPs, there should be a section on programmer productivity. There’s nothing like losing a bid because of a lack of a productive software environment to result in changes. But innovations in hardware tend to come first. Then, the software is shoe-horned in to get things to work. I refer to this as “Pornographic Programming:” You cannot define it, but you know it when you see it.
insideHPC: What motivates you? What is your passion?
Steve Wallach: When someone says “that cannot be done,” my juices flow. Too many very smart people do not get a chance to really do their thing. Startups are the major places where innovation and risks take place. My latest passion is to figure out how to use Google calendar.
insideHPC: Are there any people who have been an influence on you during your years in this community?
Steve Wallach: There are three people who have had great influence on me. One person was Ken Kennedy. The high-performance computing community lost a giant and I lost a great friend. He convinced me that a compiler could be developed that could take VAX serial code and produce vector code. And he was correct. When I came up with the idea for Convey Computer, the first person I asked for advice was Ken. I flew down to Houston and spent four hours going over the concept. At the end, he said, “This can be done, but you need a world-class compiler team.” I responded, “I know where they are.” “Where?” Ken asked. I responded: “Right where I left them.”
Another influential person was Alan Deerfield of Raytheon. Alan was a pioneer in the design of DoD specific signal processors. I learned all about: FFTs, Radar Range Gating, Kalman filtering, etc. I worked for Alan for five years in the early ‘70s. He taught me and showed me that small teams of highly motivated engineers are the most fun and accomplish the most. But you have to work hard. When motivated, working hard just comes naturally.
Lastly, Tom West of Data General showed me how to manage a group of highly motivated engineers and how to shelter the team from corporate politics. He was great at moving new products out the door, too. Tom definitely sets the gold standard for how to manage engineers. Plus, he taught me how to use high-dollar words such as “quintessential” and “canard.”
insideHPC: What “non-HPC” hobbies or activities do you have? If you ever really have ‘time off’ – how do you spend it?
Steve Wallach: I like to work out a lot – it’s kind of like training for work and keeps me mentally sharp. When I do have the time, I go to the horse track with my best buddy. I am trying to get back to my college “skill level” when shooting pool. When I can run a rack, again, I will be happy.
Also, I now have a granddaughter. Any opportunity to play with her is number one.
insideHPC: Approximately how many conferences do you attend each year? What would you say is your percentage of travel?
Steve Wallach: Let’s put it this way, in 2011, I will pass the eight-million-mile mark on American Airlines. Of course, that is not real miles traveled – that’s perhaps closer to 3.5 million real miles. And this ignores the miles on Southwest and various European and Asian airlines. My guess is that I attend, on the average, one conference a month.
insideHPC: How do you keep up with what’s going on in the community and what do you use as your own “HPC Crystal Ball?”
Steve Wallach: Attending conferences is certainly one way to see what is happening in the community. Prior to Convey, I was a contractor/consultant to Los Alamos for almost 10 years. That certainly helped me keep up. Also, I still perform due diligence for venture capitalists. Every once in a while there is a HPC type of deal, but those deals are relatively rare.
As previously described, I try to search out the HARD PROBLEMS. That is my crystal ball. And solving these problems may involve all types of technologies, hardware, software, and algorithms. I read a lot. When I find an interesting paper, I often send an email to the authors and begin a dialogue. I never know when I will use that body of knowledge.
insideHPC: What do you see as the most exciting possibility of what we can hope to accomplish over the next 5-10 years through the application of HPC.
Steve Wallach: I believe that HPC coupled with bioinformatics will lead to new ways to deal with all types of medical issues. We are already beginning to see some results. I hope one day, as described in episodes of Star Trek, we will genetically sequence a virus, take this sequence, model the behavior under certain conditions, and then synthesize a drug that hunts down the virus to destroy it.
insideHPC: What are your thoughts on HPC addressing what many are referring to as “the missing middle” which I loosely interpret as a broad spectrum of small and mid-size businesses.
Again, as I mentioned, we need an HPC App store. Any application that will aid the small and mid-size business has to be easy to use and affordable. One would also expect these applications to be available in the cloud. That is already happening. However, the user interface to these applications is different for each cloud. That slows widespread adoption.
insideHPC: What do you see as the single biggest challenge we face over the next 5-10 years?
Steve Wallach: Well, the march toward exascale computing is upon us. The consensus is that getting to exascale will NOT be as straightforward as getting to petascale. I have some thoughts on this and will be presenting those thoughts at several SC10 panels. Clearly exascale will be discussed and debated.
But the absolute biggest challenge is finding a way to get more HPC performance for less watts into our data centers. Sure, that’s related to exascale, but it goes way beyond that. If we’re ever going to solve what we used to call the “grand challenge problems,” we need some way to overcome the laws of physics that we’re facing today. By that I mean general-purpose processors just can’t get much faster because they can’t get any hotter. Today if our HPC users say “we need more performance,” we just add another 30 kilowatt rack on the datacenter floor in an attempt satisfy them.
The only way to do that is with heterogeneous computing, or to be more specific, with application-specific hardware. That means designing instruction sets that are absolutely specific to a particular application or class of applications. Which is why technology like FPGAs and GPGPUs are such a hot topic today—we’re all looking down the road and saying, “Where is this performance going to come from?”
And another thing, related to that, is that we do not need new languages. We need extensions to existing languages (like Fortran and C) that reflect the changes in computer architecture. In general, these are extensions that reflect the memory hierarchy within a node and the hierarchy among nodes. Please do not interpret this to mean I am against language research. I believe the results of this research can be reflected within current languages.
I do have one hot button in this area. I believe that MATLAB (MathWorks) is the easiest to use and the most productive HPC language. I use it all the time on my laptop. Users who are willing to accept a two-to-four reduction in performance (relative to Fortran or C) can gain an order of magnitude more user productivity. So, a great example of application-specific computing would be to build MATLAB machines to eliminate this imbalance.
In my humble opinion, cloud computing is time-sharing in a contemporary architecture. There are many models of cloud computing, thus there is no simple answer to cloud computing delivering real value. For many users and companies, having a shared resource, not having to deal with system administration and facilities is a very big win. In this area, cloud computing provides real value. Additionally, having access to additional resources for spikes in computing needs is a big win.