Distinguished Lecture Series 2009-2010: Jose Castanos

The Design of the Blue Gene Supercomputers
Jose Castanos, IBM Research

Abstract

At the beginning of this decade IBM Research announced the Blue Gene project, a new line of large scale supercomputers to attack scientific problems such as protein folding. The first generation of this architecture, Blue Gene/L, appeared in 2005. A Blue Gene/L machine at LLNL initially contained 65K processors and reached 280TFLOPS on the Linpack benchmark, becoming the fastest computer at that time. The first versions of Blue Gene/P, our second iteration, appeared in late 2007, and its largest machine, JUGENE, is expected to reach 1PFLOPs with 290K processors. IBM Research is currently designing Blue Gene's third generation.

We will begin this talk by providing a short overview of the Blue Gene software, hardware, and applications. We will then discuss the the constraints and tradeoffs we faced when designing Blue Gene's massively scale architecture, as well as our responses to expanded goals and requirements. We learned many lessons from early prototypes that had strong influence on later designs. Finally, IBM Research's culture and environment largely contributed to the success of this project so we will present some related research projects at IBM.

Biography

Dr. Castanos joined the Blue Gene project in 2000 after receiving his Ph.D. degree in computer science at Brown University. His initial assignment as a Research Staff Member involved the development of applications for high performance computing. He later became one of the technical leaders of the Blue Gene/L systems software and worked in many of its components: integrated software development environment, simulators, kernels, runtime libraries, and management infrastructure. Dr. Castanos received his undergraduate degrees in systems analysis (1988) and operations research (1989) at the Universidad Catolica, Buenos Aires, Argentina.