The IBM Blue Gene/L supercomputer at the Lawrence Livermore National Laboratory was ranked first on the TOP500 Web site in November 2005. In this paper, authored by more than 40 IBM staff members, the architecture, rollout, and accomplishments of the Blue Gene/L are presented in an easy-to-read form.
This machine uses 32-bit PowerPC central processing units (CPUs), with a special lightweight kernel on its compute nodes; this kernel performs no memory management and no scheduling. A toroidal interconnect mechanism is used, and a message passing interface (MPI) library is provided for use with standard IBM C, C++, and FORTRAN compilers. In addition to the computer nodes, there are input/output (I/O) nodes (running Linux), file servers, front-end nodes, and a service node.
The development team was based in the US, with additional members in Israel, Delhi, and other parts of India; this presented some challenges in coordination, but also provided some opportunities in round-the-clock testing. Success stories for Blue Gene/L include work in molecular dynamics, astrophysics, and biomolecular simulations. There are now more than 20 Blue Gene/L installations around the world.
The paper includes some color diagrams of system components and whole systems, and some performance charts illustrating benefits gained through the use of the lightweight compute node kernel (CNK). A useful set of references appears at its end.