Here's a collection of highlights, selected totally subjectively, from this week's HPC news stream as reported at insideHPC.com and HPCwire.
>>10 words and a link
Sun is sponsoring a Grid Engine workshop Sept. 10-12 in Germany;
https://events-at-sun.com/gridengine-workshop-2007/agenda.html
Computational Nanotechnology center opens in New York, 80 TFLOPS system;
http://insidehpc.com/2007/05/21/80-tflops-system-for-nanotechnology/
AMD publishes first Barcelona results; Intel's Penryn may pose threat;
http://insidehpc.com/2007/05/23/amd-shares-barcelona-results/
Federal University of Rio de Janeiro installs SGI for Amazon protection research;
http://insidehpc.com/2007/05/23/her-name-is-rio/
ARPANET contractor to run NSF's GENI next generation communications research facility; http://feeds.feedburner.com/~r/ComputingResearchPolicyBlog/~3/118488140/000598.html
QLogic Ships Ethernet and Fibre Channel Card for BladeCenter;
http://insidehpc.com/2007/05/23/qlogic-ships-ethernet-and-fibre-channel-card-for-bladecenter/
>>Server market: we're number 1. And so are we.
Gartner and IDC release server sales survey results this week and show IBM and HP are number 1.
Market analyst firm IDC just released its Q107 Worldwide Quarterly Server Tracker. That report shows that HP shipped more servers than any other company, and lead the HPC market again in the first quarter of 2007. This is the 20th consecutive quarter that HP has led industry server sales. From HP's release, “In the high-performance computing market, HP maintained its No. 1 spot for the first quarter of calendar year 2007 with 33.7 percent revenue share.” More details are at http://insidehpc.com/2007/05/24/idc-hp-leads-hpc-market-in-q1/.
And just to make sure you have a hard time identifying who's really out front, Gartner announced today that IBM is number one in server revenue share. The difference between this announcement and the HP one appears to be that IDC shows HP as leading in units shipped, while Gartner shows IBM as taking in more money. If I'm right, I'd rather be IBM. More on IBM's results at http://www.hpcwire.com/hpc/1581037.html.
>>FY08 appropriations: DOE Office of Science doing well
The CRA's Policy Blog gives us a peek at the first FY08 appropriations numbers, and the DOE is doing well: “Though we don't yet have all the detail about increases in individual accounts, we do know that the Office of Science would see an overall increase to $4.516 billion in FY 2008, which is $120 million above the President's request for FY 2008 and $719 million above the FY 2007 level, or an increase of 18.9 percent.”
While we don't yet know how the money is going to get divided within the department, this could mean some increase for the Advanced Scientific Computing Research Program. We'll know in June when the full committee marks up the bill.
The CRA's post has links to the numbers at http://www.cra.org/govaffairs/blog/archives/000599.html.
>>Profile on Sedna
Sedna is Sun's proximity interconnect project; the man behind the implementation is Hans Eberle. You may recall that the Big Idea behind proximity communication is to transfer data between chips without using wires. Eberle is already reporting results of I/O bandwidth two orders of magnitude higher than by conventional methods. From the profile:
“Instead of using a complex, multi-stage, hierarchical design where you actually have to schedule the path through the various switching elements, requiring a fantastic degree of coordination, the new design is a simple, single-stage switch. 'With Proximity Communication we're getting something like two orders of magnitude more I/O bandwidth. What that means is we don't have to make use of hierarchical topologies. We basically can look at this as a flat switch,' Eberle says.” You can read the rest at http://research.sun.com/minds/2007-0518/.
>>IBM makes Power6 available June 8, shares results
The Register opened the week with some actual Oracle performance numbers. Oracle pulled the benchmark results off their website when they noticed the story. Here's an excerpt:
“The Register has spotted four 4.7GHz — yep, you read that right — Power6 chips cranking on Oracle 11i… With 4.7GHz chips (4MB of L2 and 32MB of L3 cache), an IBM p570 server showed an average response time of .625 seconds when handling requests from 2,100 users. That compares to a p570 with 2.2GHz Power5+ chips that had a response time of .983 seconds for 2,000 users.” The full article can be found at http://www.theregister.co.uk/2007/05/20/ibm_power6_oracle/.
This week PC World also reported IBM's figures that show the Power6 three times faster than the Itanium used in HP's Superdome, with 30 times the bandwidth. The Power6 is IBM's first chip that does decimal math in hardware, possibly giving it index calculation advantages over other offerings (for fast database traversal), and of course IBM is claiming the obligatory “green” cred for the chip. You can read more about it at http://insidehpc.com/2007/05/23/ibms-power6-in-the-wild/.
>>Building a cluster with Windows CCS?
There are oodles of information sources out there to help get you started, courtesy of Microsoft's influence as the 9,000 pound gorilla in IT. But two new ones showed up in this week's news stream, and I thought I'd point them out to you. The first comes courtesy of Fujitsu in the form of a “best practices” paper on building a Windows CCS cluster with their servers and switches, Myricom cards and drivers, and Active Directory. From the company:
“The Best Practice paper is intended to provide the fundamentals of configuring Fujitsu 10 GbE switches, as well as installing software for setting up a complete cluster. Tested in the Microsoft computing labs, this best practice installation approach will further service the industry, educate IT professionals on proper deployment and implementation guidelines, and facilitate the installers' success.” Read more about it at http://insidehpc.com/2007/05/21/building-a-cluster-with-fujitsu-and-ccs/.
Dan Fay also points to a paper in The Architecture Journal detailing the configuration of an HPC cluster using Windows CCS and a bunch of other MS technology for a solution that focuses on workflow. From the article: “Microsoft Cluster Compute Server Edition is easy to include as a service gateway inside a general n-tier application structure, and is simple to integrate via command-line or API hooks.” More information can be found at http://insidehpc.com/2007/05/24/466/.
—–
John West summarizes the headlines in HPC every day at insideHPC.com, and writes on leadership and career issues for technology professionals at InfoWorld and on his own blog at onlytraitofaleader.com. You can contact him at [email protected].