Posted on 02/04/2005 8:33:26 PM PST by Swordmaker
The new Turing Xserve Cluster at the University of Illinois at Urbana-Champaign (UIUC) is the one of the latest large-scale high speed computing systems to use Apple's Xserve G5. The director in charge of the program said the decision to use Apple hardware and software made sense based on price, performance, compatibility, efficiency and support.
The Turing Xserve Cluster -- named after Alan Turing, the famed British mathematician credited with founding the field of Computer Science -- is a 640-node system that uses 2GHz dual-processor Xserve G5s, Apple's rack-mounted server. Running Mac OS X v10.3 Server, the systems talk using Myrinet's Myricom networking hardware. In addition to local storage on each Xserve, the system also uses an Apple Xserve RAID system that provides a total of seven terabytes (TB) of storage.
UIUC's Computational Science & Engineering (CSE) Program plans to use the Turing Xserve Cluster to do the heavy lifting for the University's computational research efforts, according to Michael Heath, Director of Computational Science and Engineering for UIUC.
"We needed as much capacity as we could get," Heath told MacCentral. "We do very large scale simulations for rockets, for example. It's a first-rate place to do computational research, and we wanted a first-rate environment in which to do that."
Heath said Apple ultimately won out over other competitors based on the Xserve's performance, affordability, and a number of other features the University felt was important. Heath estimates the total cost of the Turing Xserve Cluster at around US$3 million.
Speedy, cool and efficient
The Turing Xserve Cluster replaces a previously installed 208-node Linux-based system. Heath told MacCentral that the new Xserve-based system provides solid benefits over its predecessor. "We're seeing a tenfold increase in computing power," said Heath.
Another attractive feature was the Xserve cluster's power and cooling requirements. "We already knew we were going to have to upgrade our computer room to accommodate the new systems, but Apple's Xserve solution was more efficient by a factor of two than other competitive systems," said Heath.
Heath is reluctant to provide specific benchmarks for the Turing Xserve Cluster at this point. "It's a little early to do things like run a Linpack benchmark," said Heath, referring to a method used to measure the floating-point speed of a computing system. Linpack benchmarks have been used over the years to gauge the performance of high speed clusters and supercomputing systems.
Heath added that a "back of the envelope computation" yields a theoretical performance of about 10 teraflops, though he admits the system is working well below that level right now. "There are a lot of optimizations that we can do yet," Heath added. "But the system is still too new and it's too soon to have done that."
'The Ultimate in Compatibility'
UIUC put together the system in just a couple of months. Though renovations to the computer room where the Turing Xserve Cluster took longer than the University expected for reasons unrelated to the Xserve, the room was ready for the new hardware in late November. Preliminary testing occurred in January, and Heath said that the Turing Xserve Cluster has been running since then. "... Though we're just getting our first real users now," he said.
UIUC has a history with Apple. Qualcomm's Eudora -- one of the first Internet e-mail clients for the Macintosh -- got its start there in the late 1980s, the work of Steve Dorner. UIUC is also the home of the National Center for Supercomputing Applications (NCSA), where Marc Andreessen and Eric Bina created Mosaic, the first graphical Web browser -- available almost from the start for the Mac, and laying the groundwork for what would ultimately lead to Netscape Navigator. This long connection with Apple certainly worked in the company's favor when it came time to evaluate a new computing cluster, according to Heath.
"There's certainly a comfort level with Apple borne of a number of years of experience," Heath said, adding that he uses a desktop Macintosh and a PowerBook. "Having a cluster system based on Macs is the ultimate in compatibility," said Heath.
Also working in the Xserve's favor is Mac OS X's basis as a Unix operating system. "The learning curve is much shorter than it would be otherwise," said Heath. "Our users are all used to various flavors of Unix, so coming to grips with Mac OS X Server is totally natural for all of them." Heath added that Apple's support for industry standards and use of open source components also factored positively into UIUC's decision to go with the Xserve.
Heath said that he couldn't be happier with Apple's support before, during and after the installation process. It's like night and day compared to the old system.
"You know the Linux world," said Heath. "You are totally on your own. You're getting your operating system and hardware from at least two different sources. This is much more integrated."
Apple is participating in weekly conference calls with the University to make sure the Turing Xserve Cluster is working well and meeting the center's needs. "We've been very impressed with their continued interest and involvement," said Heath.
Another Mac Supercomputer... PING!!!
If you want on or off the Mac Ping list, Freepmail me.
I want one
It followed me home
Can I keep it pleeeeeeze??
Somebody from the left has hacked CYNDISLIST.COM
CYNDISLIST.COM (Genealogy) ^ | 2/4/05 | Unknown
Posted on 02/03/2005 9:44:30 PM PST by FUMETTI
http://www.freerepublic.com/focus/f-news/1335633/posts
That's when you get into the serious supercomputer hardware. The backplane has to be wide (128bits prefer 256) and interleaved. That's what makes a Cray a Cray.
How does this set-up compare to what Cray offers - apples to apples?
It doesn't. It's an apples to oranges kind of thing. But when I hear Apple folks say they have the fastest super computer I like to remind them. FWIW - I go back with apple to Lisa days.
OK..thanks for the follow up.
Yeah - it could have been a nasssty little machine for clustering if they's only given it a gigabit ethernet chipset. How much more could it have added to the price?
504 CPU Cray X1 = 5,895 GFlops, or 11.70 GFlops per CPU.
2200 2.3MHz G5s = 12,250 GFlops, or 5.568 GFlops per CPU
My 1.8 MHz G5 = (RAverage) .27 GFlops, if my ~ 4 Hrs per 3.87 TFlop Seti@Home work unit processing speed is any indication.
I feel so slow.
Have you thought of using IP over Firewire?
You might still be better off waiting for G5 minis.
There are a whole bunch of other reasons not to.
University of Illinois AND terabyte POING!
SEVEN terabytes actually...could you handle it?
Disclaimer: Opinions posted on Free Republic are those of the individual posters and do not necessarily represent the opinion of Free Republic or its management. All materials posted herein are protected by copyright law and the exemption for fair use of copyrighted works.