Sun Microsystems today announced the availability and some additional offerings for its high performance computing (HPC) product line. The news, issued at the Supercomputing 2007 conference in Reno, Nevada, builds on the previously announced Constellation HPC clustering.
Sun announced Constellation back in June. Constellation is a three-pronged petaflop-capable system consisting of ultra-dense blades, Sun's X4500 storage systems and the new, ultra-dense Magnum switch. The blade system is known as the Sun Blade 6048, with 48 blades in four shelves of 12.
Magnum is the codename for an Infiniband switch with 3,456 ports. The official title is Sun Datacenter Switch 3456. Sun claims its architecture will greatly reduce latency when communicating from node to node and offer improved overall system throughput.
The first Sun Constellation System is currently being built at the University of Texas in Austin's Texas Advanced Computing Center (TACC). This system will consist of 82 blades, the Sun Fire 6048 Ultra dense blade server, 72 Sun X4500 storage servers and two 3456 switches.
Sun and TACC are building the cluster for use by the National Science Foundation, so it will be a part of the TeraGrid, a national-scale cyber-infrastructure facility funded by the NSF. "This will more than double the compute power on the TeraGrid," Bjorn Andersson, director of HPC and integrated systems at Sun told InternetNews.com. "It will be used for the kind of science done on the TeraGrid today but on a larger scale."
Sun is also taking the wraps off of the StorageTek 5800 system for large-scale storage of fixed data. Sun claims it will be the only open source object storage system today (the StorageTek 5800 system sits between the X4500 and the tape backup system).
Andersson said the 5800 is meant to be a middle ground between tape backup, which is mostly writing and a little reading, and high speed server storage like the X4500, which is as much writing as reading of data. "It's read many times and writes a few," he said.
The Constellation system is very mix and match. Customers can choose between Intel Xeon, AMD Opteron and Sun Niagara 2 UltraSparc processors and can run Solaris, Linux or Windows Server, in any combinations.
"We're making cluster computing much more accessible and scalable than it has been in the past," said Andersson. "We're working on scaling up and down and making sure we can start with a rack and scale all the way out to the two petaflop level."
The two petaflops is a theoretical limit and that's at peak performance, not sustained. TACC will max out at 521 teraflops, which would put it well ahead of the fastest supercomputer, IBM's massive BlueGene/L system at the Lawrence Livermore National Labs, which peaks out at 360 teraflops.
Of course, that list is subject to change. Top 500, the group that maintains a list of the top 500 fastest supercomputers, will issue an updated list at the supercomputer show this week.
TACC, if complete now and at peak performance, would easily surpass BlueGene/L and be only the second Sun entry on the list. The other, #52, is a cluster at the University of Southern California. But Andersson insists Sun isn't trying to win a race.