Imagine a server with 110 Terabytes of hard disk. That’s roughly what the National Center for Supercomputing Applications, better known as NCSA, has developed.
NCSA, a government-funded research center at the University of Illinois at Urbana-Champaign, has been on the frontiers of computing since it was founded in 1986. NCSA Mosaic, the first graphical Web browser, for example, emerged from its doors in 1992.
The center’s latest creation is a massive storage area network (SAN), which is hooked directly to 256 separate computers. The entire 70 terabytes of disk — slated to grow to 110 terabytes this summer — is made up of one single file system, so each one of the computers can address the entire amount of storage.
That is pushing the envelope of storage technology, says Michelle Butler, NCSA’s technical program manager responsible for storage. “We are on the bleeding edge,” she says. “No where else in the world today can you hook up 256 Linux machines [to this much storage], and have each of them see the exact same data as the machine right next to it.”
Among other things, this provides an enormous performance boost, says Butler. The 256 computers — all Intel-based Linux servers from IBM — are clustered together. Because the storage for the cluster is one single file system, all 256 machines can write to the same file in parallel. “Data gets written to disk 256 times faster,” says Butler, “because they’re all doing it at once.”
The system also gets a speed boost because the disks share data directly over their Fiberchannel connections, rather than going out over a network. NCSA is using Brocade 12000 network switches and storage technology made by LSI, and purchased from IBM.
As if this wasn’t enough storage, later this summer NCSA will be installing another 170 terabytes of storage, which will serve a second cluster of 768 Linux machines.
This huge SAN is part of an even larger experiment in large-scale computing in which NCSA and five other research facilities around the country are creating a massive storage- and compute grid which will have more than 1 petabyte (1 quadrillion bytes) of storage capacity. Called TeraGrid, the grid is being built by NCSA along with supercomputing centers at Carnegie Mellon University, the University of Pittsburgh, the University of California at San Diego, Argonne National Laboratory, and the California Institute of Technology.
A New Type of Science
The high-performance 110 terabyte storage system NCSA is creating is doing more than just pushing the envelope in computer science, says Butler. It’s also enabling a whole new class of research projects by the scientists who use NCSA’s computing facilities.
Supercomputing in the past has focused on CPU-intensive computing, according to Butler. “The supercomputing world has been taught over the years not to do data-intensive tasks, because I/O is slow, so your CPU cycles are sitting there spinning while I/O is going on,” she says.
Now, however, NCSA is starting to see a new class of more data-intensive applications. Researchers trying to predict the weather, understand how black holes work, or do genetic sequencing are starting to use NCSA’s computing power to crunch large quantities of data, says Butler.
So are some of NCSA’s industrial partners, which are commercial companies the research center works with to ensure that its research is meeting the needs of the private sector as well as pure researchers.
One heavy equipment manufacturer, for example, was able to use NCSA’s visualization resources to model the cab of one of its new machines. That led to the realization that design prevented someone sitting in the driver’s from reaching all the levers needed to operate the equipment.
“Having all this storage available is enabling us to do this kind of science, which is not something we could do in the past,” Butler says.