High-performance computing veteran Thomas Sterling has the respect of his peers and the industry. His excellent speaking skills, candor and humor complement his HPC expertise and public persona. He is also admired by young researchers and students - they seek him out at scientific computing conferences for advice on research projects and career choices. This year, at ISC'13, Sterling will be delivering a keynote on HPC Achievement and Impact - 2013, thus marking a decade as an ISC keynote speaker. Here, he is speaking to ISC's PR manager, Nages Sieslack…
Please tell us what drew you to the field of high-performance computing?
It was an incremental process. When I first matriculated at MIT as a graduate student (1977) the Cray-1 was still the fastest computer in the world and microprocessors were 8 bit architectures with early 16 bit chips (e.g., Intel 8086) coming down the pike. In my second year, I was challenged by my faculty advisor to pursue the question of applying these new emerging technologies to the simulation of power electronic circuits like switching regulators; an area that still made heavy use of analog computers (bunches of op amps solving first-order differential equations). I built a machine of 8 worker processors and a master in an S100 bus based chassis and ran it in lightweight SPMD mode. It never worked well but it exposed me firsthand to many salient issues associated with incipient multiprocessors. My Ph.D was on the deeper question of parallel paradigms and execution models before it was clear what that even meant. We built a shared memory machine at MIT called Concert (led by Bert Halstead) and dealt with many of the communication issues along with programming and synchronization questions. Slowly - very slowly - I started to build up some insight into the foundational challenges and models that dictated parallel computing. I've been pursuing it every since. More than 30 years later, I think I've almost got it; any day now.
You are known as the father of the Beowulf cluster, but you've continued to pursue other forward-looking projects and ideas in high performance computing. What would you like people to remember you by?
My name. But beyond that, I think my principal strength and contribution has been to consider the challenges that the field has faced throughout its rapid evolution. Through a maturing and deep understanding of underlying principals of execution model and performance model, I continue to explore the cross-cutting design space of hardware, architecture, software, and programming methods. This is a time of transition from principally static methods to dynamic, adaptive control requiring advanced strategy and paradigm. It would be presumptuous of me to assume that I will be remembered at all. But if you will forgive the potential hubris and if current research proves favorable, I would not be unhappy if the ParalleX execution model makes a recognized contribution to the future of our field. I say this with full acknowledgement of all who have contributed to many of its underlying precepts as well as those who are contributing to its realization.
You've lived through the entire supercomputing era and, in some cases, been at the center of it. Have you ever considered writing a book about this experience?
EDSAC and I were born in the same year. At age 63, supercomputers have extended their performance capability by a factor of ten trillion or more through a process of technology-driven reinvention at least six times through as many decades. What is required right now is not a retrospective but a forward-looking guide book to imbue the emerging generation with the knowledge, concepts, and skill set essential to empower continued development and exploitation of current and future generation high performance computing. I am developing a textbook for that purpose with my colleagues Andrew Lumsdaine and Chirag Dekate based on a course currently being taught at Indiana University, and years of experience with a similar course previously at LSU. There is also an important need to share with the citizenry of the US and the world the essential association that supercomputers have with humanity's daily life and future. Why is it that a high school student is more likely to know about the Hubble telescope, the common microscope, or even "atom smashers" than they are to understand the continuous impact on them of supercomputers? I think two things: we don't teach about supercomputers in K-12 [primary and secondary education in the US] and there is no widely disseminated explanation that is accessible and indeed exciting to the public. At LSU we conducted (they still do) the annual week-long 'Beowulf Boot Camp' that attracts high school students from around the state of Louisiana to immerse themselves, hands-on, with Linux clusters. We need to craft a handbook that can serve as a guideline for such events. To address the second challenge, I've been outlining a book for the general public entitled "Hammer of the Mind" (no publisher yet) that will try to instill the sense of excitement and appreciation for the transformative effect of supercomputers on the present and future. While I have been involved in a number of successful book projects having co-authored six over the last couple of decades, each one is different and these two are entirely new ventures. Let's hope I can complete these; they would both serve as contributions. We certainly need them.
You are known for your dry wit and demeanor in a field that rarely attracts such people. What, or who, do you attribute your sense of humor to?
This is awkward. Nobody loves a clown; they only love to look at them. I'm not sure I would concede the point. I certainly don't know how to tell a formal joke. Instead, I am truly entranced by our field, the intellectual challenges and the really remarkable people who engage them. We so easily forget that what we are doing is truly unique in the history of humankind; that this is as much about the human experience as it is the technology that is a part of it. And as in any human endeavor, it comprises the hopes, dreams, and aspirations of people as well as the ironies, serendipities, joys, and disappointments that are essential components of the human condition. Humor, the ability to instantly detect and appreciate contradictions to the norm, is at its core an element of what it is to be human. Our field is permeated with events of this type. My only ability and inclination is to reflect and share my perspectives of such events and conditions with our colleagues. I am not so sure I have a sense of humor as I do an intrinsic tendency to delight in the humanity of the machines we create and employ.
What's the most inspiring thing you see happening in the field of supercomputing today?
At the risk of appearing self-serving, I am really excited about what I perceive as this period of transition between the paradigm of the past and the execution model of the future. This viewpoint is not widely held but I am convinced this it is what we are seeing. We don't really have a choice. Technology demands it as it has many times in our short history. Indeed, for practical reasons of markets, road maps, and legacy codes we have deferred this too long. Each time this has happened, we find new ways to address in synergy the fundamental problems that have always faced parallel computing: starvation, latency, overhead, and the waiting due to contention for shared resources. Now we need to find ways to exploit the exponential growth of numbers of cores and the heterogeneous mix of their internal structures and external organization. It is likely that we will tap the largely unused runtime information combined with adaptive methods to significantly improve local efficiencies and vastly expand global scalability. After the brilliantly successful Pax-MPI era of the preceding two decades, we may be reformulating the interrelationships across the system layers in a transformative approach embracing dynamic adaptive cooperation and control. Such periods of change in our field are rare and it is inspiring to be a part of it.
How about the most frustrating thing?
It may be a small thing, but there is a friction within our community among those engaged in developing future paths and a resistance to those exploring alternatives to conventional practices and their incremental extensions. I find this a bit frustrating. We should encourage innovative thinking and be prepared to allow the science of computer science to determine the ultimate outcomes for the future of our field. Such investigations should be acknowledged as constructive, not disruptive.
Finally, when you're not thinking or talking about supercomputing, what do you like to do to relax?
I like to sail. It's about the only time I don't think about work. I also like to read history and biographies. My favorite period is the late Bronze Age. Finally, I have had a deep interest in astronomy and cosmology from childhood and I enjoy reading about advances in these related fields.