Oklahoma
Supercomputing Symposium 2003
Wednesday September 24 -
Thursday September 25 2003
at the University of Oklahoma
Dr. Jose G. Castanos
Research Staff Member
IBM Thomas J. Watson Research Center
Dr. Jose G. Castanos obtained his undergraduate degree in Operations
Research at the Universidad Catolica Argentina, and his Sc.M. and
Ph.D in Computer Science at Brown University. In 2000 he joined the
Blue Gene team at the IBM T. J. Watson Research Center to develop
and analyze the initial molecular dynamics codes for this architecture.
He has worked on a wide variety of systems software and performance
analysis projects related to Blue Gene. He developed the multiprocessor
simulation infrastructure, collaborated on compilers, kernels and
libraries, provided a software development environment for Blue Gene,
designed its control system, and studied applications and benchmarks on
this architecture.
Talk Abstract
Blue Gene/L: Delivering Large Scale Parallelism
In this talk we present an overview of the hardware and software
architecture of the Blue Gene/L Supercomputer, a jointly funded
research partnership between IBM and the Lawrence Livermore National
Laboratory. This massively parallel system of 65,536 nodes is based
on a new architecture that exploits system-on-a-chip, low cost,
embedded Power PC technology to deliver target peak processing power
of 180/360 teraFLOPS (trillion floating-point operations per second).
The target value of 360 teraFLOPS peak is appropriate for applications
that can utilize both embedded PowerPC processors on a node. For
applications that require significant messaging protocol overhead, one
of these processors may be dedicated to this messaging and is not
available for computations. For this later class of applications, the
target peak performance is 180 teraFLOPS. The machine is scheduled to
be operational in the 2004-2005 time frame (early prototypes are
already available), at price/performance and power consumption/
performance targets unobtainable with conventional architectures. The
large parallelism of this machine represents a new level of complexity
Slides:
PowerPoint
PDF