Edgar Leon, a computer scientist in the Livermore Computing Division, has spent his career confronting daunting challenges in the development of ever more powerful high performance computing (HPC) systems. He helped prepare for Sequoia, the 17.5 petaflops (quadrillions of operation per second) IBM Blue Gene/Q supercomputer ranked No. 3 in the world on the Top500 list, and now is laying the groundwork for Sierra, the 125-petaflop IBM system scheduled for delivery in 2017.
Working on world-class supercomputers at a U.S. national laboratory was not what Edgar, a native of Mexico, envisioned when he began preparing for university. With strong family encouragement, he initially set out on a path to medical school. But as part of his high school preparatory studies, he earned an associate degree in computer science that would turn out to be fortuitous.
“I was intrigued by programming languages and how they could be applied to solve a variety of problems,” Edgar says. “Curiosity about why we do things a certain way is part of my personality and computing provides a vehicle to satisfy such curiosity.” He began his undergraduate studies in computer science at UNAM, the largest university in Mexico, and completed his senior year at the University of New Mexico in Albuquerque, where he also earned his master’s and doctoral degrees in computer science.
Today, Edgar is a senior member at the Institute of Electrical and Electronics Engineers (IEEE), a prestigious status that reflects his professional maturity, extensive experience, and documented achievements. His research at LLNL is part of a larger U.S. Department of Energy effort to develop and leverage exascale computing capabilities to ensure the nation’s global leadership in HPC. In particular, he works on performance and reliability of communication libraries at scale, emerging memory technologies for exascale systems, power-aware computing strategies, and resilience of future systems.
The principal challenge is ensuring the “resilience and reliability” of complex HPC systems with tens of millions of components that operate under great power constraints, Edgar says. “The more components you have, the more failures occur. A key question in deploying state-of-the-art supercomputers is how our applications can take advantage of these new capabilities.”
Edgar's pioneering contributions include a mechanism to simulate novel computer architectures in large systems and the application of “cache injection” to HPC. The latter contributed to the commercialization of cache injection, a technique for tolerating memory latency, or slow data access, in HPC systems.
Edgar finds his LLNL assignment profoundly satisfying because it provides a balance of applied scientific research and working with people. “Here at Lawrence Livermore, we’re driven by missions in the national interest and not by commercial products, yet we work with top people from academia and industry as well as other laboratories,” he says.
Recalling his days as a student and the mentors who helped him in his career, Edgar dedicates time and energy to mentoring and supervising students. “The challenge for a mentor is to identify and leverage the unique skills each student brings to an internship. Not everyone is the same,” he says. “Students are excited by the magnitude of the problems they work with here and, at the same time, we can have an impact on the focus of their studies and their dissertations.”
“Our impact on people is as important as our achievements in science,” he adds.