New SGI supercomputer to scale Linux to 1,024 CPUs
The National Center for Supercomputing Applications will use it for research
Silicon Graphics Inc. is building an Altix supercomputer for the National Center for Supercomputing Applications (NCSA) that will run a single Linux operating system image across 1,024 Intel Corp. Itanium 2 processors and 3TB of shared memory.
Rob Pennington, interim director of the NCSA, which is based at the University of Illinois Urbana-Champaign, said the new machine will be very different from the existing machines at the facility, which include several Linux cluster supercomputers. Until now, the largest shared-memory supercomputer available to scientists there was an IBM p690 machine with 12 32-processor nodes.
With the new Altix machine, researchers will have far more computing power for their work, which includes weather-data analysis, simulations of black-hole collisions and other large-scale events in the evolution of the universe.
Earlier cluster supercomputers at the NCSA used multiple images of the Linux operating system -- one for each node -- along with dedicated memory allocations for each CPU. What makes this system more powerful for researchers is that all of the memory will be available for the applications and calculations, helping to speed and refine the work being done, Pennington said.
"The users get one memory image they have to deal with," he said. "This makes programming much easier, and we expect it to give better performance as well."
Initially, Pennington said, the system will use two images of Linux -- one per 512 processors -- while it's being tested and configured. Later, all 1,024 processors will address one image of the SGI Advanced Linux operating system being used. That operating system is based on Red Hat Enterprise Linux.
Dan Kusnetzky, an analyst at market research company IDC in Framingham, Mass., said the Altix system follows a path of innovation that SGI has offered for years in the supercomputing market. "SGI has often led the field in how many processors they could run on one operating system," he said.
The system, which is being called Cobalt, is a symmetric multiprocessor machine that will be connected to a 370TB SGI InfiniteStorage shared-file system, according to Mountain View, Calif.-based SGI. The storage will also be accessible to the other supercomputers at the NCSA.
The construction of Cobalt began with the delivery of the storage equipment last month, and the machine is expected to be fully online by March 1. It has a potential peak performance of more than 6 trillion floating-point operations per second (TFLOPS), which will bring the total computing power at NCSA to more than 35 TFLOPS and disk storage to three-quarters of a petabyte.
SGI uses a proprietary NUMAflex shared-memory architecture that allows memory to be shared across multiple commodity processors and SGI ProPack for Linux, an application package that allows Linux to scale to larger requirements, according to a statement from the company.