Q&A With IBM\'s Blue Gene/L Chief Architect

Edison

Dr. Alan Gara is the chief architect of the IBM Blue Gene/L, the world's most powerful supercomputer. Dr. Gara also led the design and verification of the Blue Gene/L compute ASIC as well as the bring-up of the Blue Gene/L prototype system. His designs are noted not only for performance, but also for innovative architecture that consumes much less power and floor space when compared to other supercomputers. A team from Lawrence Livermore National Laboratory and IBM Research, including Dr. Gara, were awarded the Gordon Bell Prize for leveraging Blue Gene/L performance. Last week, he was named an IBM Fellow, the company's most prestigious technical honor.

We got the opportunity to ask Dr. Gara about the Blue Gene technology and some of his thoughts about the nature of supercomputing.

HPCwire: What was the problem Blue Gene addressed?

Gara: Both the cost and the power of supercomputers have reached unprecedented and unsustainable levels. This is inconsistent with the necessary growth in performance needed to address future performance requirements. Current machines either utilize expensive custom hardware to address hard to scale problems or they utilize commodity processors with commodity networks which do not scale very well. Neither of these approaches is power efficient. Advances beyond the state-of-the-art requires innovations that address the fundamental constraint of power efficiency while improving cost performance and system usability and scalability.

HPCwire: How does Blue Gene address this problem?

Gara: Blue Gene/L's combination of high performance with smaller size, low cost and low power consumption has brought supercomputing technology to the point where it can now be made more widely available and applied to a broader set of applications. In creating a new paradigm of supercomputers optimized for massive parallelism, the new architecture offers an unprecedented scalability and an ability to handle large amounts of computation while consuming a fraction of the power and floor space required by today's fastest systems. This discovery is helping to define the path for future computing and will allow for the simulation of physical processes that will lead to scientific discoveries on many fronts, accelerating progress in a range of fields including life sciences, hydrodynamics, materials sciences, quantum chemistry, quantum physics, molecular dynamics and business applications.

HPCwire: Please share an anecdote about a challenge that needed to be overcome in building the Blue Gene architecture?

Gara: Software has often been the Achilles heel of supercomputers. One of the most challenging aspects of the Blue Gene/L system design has been the development of software that can be scaled to the unprecedented levels of more than a hundred thousand processors.

In designing the Blue Gene/L system software, we followed three major principles: simplicity, performance and familiarity. Because we targeted, Blue Gene/L primarily for scientific computations, we kept the system software simple for ease of development and to enable high reliability. For example, we impose a simplifying requirement that the machine be used only on a strictly space-sharing mode -- only one (parallel) job can run at a time on a Blue Gene/L partition. Furthermore, we support only one thread of execution per processor. Another major simplification is the preclusion of demand paging support in the virtual memory system, this limiting the virtual memory available per node to the physical memory size.

These simplifications lead directly to performance benefits that allow us to take advantage of hardware features and deliver a high-performance system with no sacrifice in stability and security.

HPCwire: What's ahead on the Blue Gene technology roadmap?

Gara: IBM remains committed to building a petaflop computer as announced in 1999 when the Blue Gene project was initiated. There has been lots of activity in this direction since that announcement. Blue Gene/L is a significant step towards this goal.

We will of course not stop at a petaflop. There remains enormous scientific challenges that require multi-petaflop scale computing.

HPCwire: What is your hope/vision for supercomputing in the next ten years?

Gara: I think that the next ten years of supercomputer will be an incredibly interesting time. Most areas of research go through periods of intense change followed by times of slower evolutionary growth. In the area of supercomputing we are entering a period of intense activity and change being forced upon the industry because of the disruptions happening in silicon technology. Future silicon technology offers higher density without much power efficiency gains. This is a fundamental shift that will cause everyone to rethink how to architect supercomputers. We either need to do this rethinking or we will not be building any more powerful machines. We already have supercomputers in the tens of MegaWatt range. We will want computers to be hundreds of times faster which would mean GigaWatts of power if we don't rethink the conventional wisdom. Clearly we will not be building GigaWatt scale computers because of the cost implications.

HPCwire: Who or what inspired you to become a computer scientist?

Gara: Interesting question. My career has evolved in way where it is difficult to characterize one discipline. I received my degree in theoretical physics. I have also always been passionate about making "real" things and have been involved with electronics since I was a child. My electronics interests allowed me to segue into experimental physics followed by an interest in building supercomputers to solve theoretical physics problems.

The Blue Gene/L project presented challenges that extended beyond electrical engineering and would be considered computer science. So the team which also has a somewhat unconventional background, had to become somewhat expert in computer science. Along the way I have has the tremendous advantage of collaborating with world leaders in each of these different aspects and I believe it is really the combination of these different disciplines that makes the Blue Gene team unique.

HPCwire: What advice would you give to someone interested in this field?

Gara: I guess I'd say that certainly surrounding yourself with amazing people is one part of the equation; it certainly worked out well for me. Also, with these people, everyone's different and one needs to learn the strengths of all the people that you're working with and also listening to those people and trying to find the best path that's consistent with all those constraints I think is really the key.

http://www.hpcwire.com/hpc/677493.html