Düstere Aussichten für Netzwerkspezialisten? Müßen Heerscharen von IT-Spezialisten sich um ihren Arbeitsplatz sorgen, weil es so einfach ist, Macs zu leistungsfähigen Clustern zusammenzuschließen. Bisher dominieren hier Pentiumrechner, die sich der Geheimwissenschaft Linux bedienen, welche so kompliziert ist, daß sie nur von einem exklusiven Zirkel beherrscht wird, der sich seine Kenntnisse teuer bezahlen läßt. Mit Macintosh-Rechnern wird dieser Prozeß kindereinfach und eine ganze Berufsgruppe muß um ihre Zukunft bangen. Dieses Szenario zeichnet zumindest folgender Artikel von wired.com, mal sehen wie's wirklich kommt. Auf jeden Fall zeigt dies einen Weg auf, wie die Strategie, in den Bildungsbereich zu investieren, aufgehen könnte.
Grüße Max
That's a Whole Lot of Power, Mac
By Leander Kahney
www.wired.com
Several Macintosh computers are networked together for incredible computing power.
2:00 a.m. Jan. 29, 2002 PST
Unlike the new iMac, Apple's new PowerMac doesn't have a redesigned enclosure. But it does have a pair of G4 chips that break the gigahertz barrier for the first time, a psychological hurdle that is making Mac fans drool.
Unveiled on Monday, the PowerMac features G4 PowerPC chips from Motorola that run at 1 GHz. Apple also introduced lower-end models with chips that run at 800 MHz and 933 MHz.
The high-end machine also debuts Nvidia's new GeForce4 graphics card and a "SuperDrive" that can burn DVDs and CDs. Available in February, the new PowerMacs cost between $1,600 and $3,000.
Although the new PowerPC chips are just half as fast as processors from Intel and AMD, which are now running at up to 2 GHz, many think the chip is still more powerful, thanks to its unique architecture.
Apple promises the dual-processor machine can perform an "amazing" 15 billion floating-point operations per second, or 15 gigaflops. According to Apple, this allows the machine to run Adobe Photoshop about 70 percent faster than an Intel Pentium 4 at 2 GHz. And encoding video is 300 percent faster, Apple claims.
At 15 gigaflops, the new PowerMac is firmly in supercomputer territory. When Steve Jobs unveiled the first G4 PowerMac two years ago, the 500-MHz chip performed at up to one gigaflop.
It was so fast that it was classified as a weapon and couldn't be exported to countries such as China, Iraq and North Korea. The new machines don't suffer from the same export problem; the Bush administration has effectively removed the restriction.
The new PowerMac is stimulating a lot of people's techno lust, including the nerds at Slashdot, who until recently were cool on Apple. But Mac OS X, which is based on Unix, is steadily drawing Linux aficionados into the Macintosh fold.
The new PowerMac is especially attractive to scientists who build supercomputers from clusters of Macintosh machines.
"It looks like dynamite," said Victor Decyk, a physicist at University of California, Los Angeles, who recently helped to build the largest Macintosh cluster yet, by hooking together 56 dual-processor G4s. "I'm going to order one as soon as I can."
A few years ago, Decyk and a pair of colleagues began playing around with G3 Macintoshes and were impressed with their performance.
"Not only was the performance faster than the Pentiums but it was comparable to the performance achieved on some Crays," the team said in a report.
Further investigation showed that Macs are very easy to hook into parallel clusters and perform extremely well, thanks to the PowerPC chips and Mac OS X.
Dean Dauger, one of the team members, recently got an 8-node Mac cluster to perform an ultra-complex calculation that contains 100 million mutually interacting particles. A few years ago, the same calculation could only be performed on the world's largest supercomputers, he said.
Clusters are becoming an increasingly common way to perform supercomputer tasks on the cheap. Simply hook up a bunch of off-the-shelf computers and set them to work in parallel on complex problems. Most clusters are based on Pentium machines that run Linux. But according to Dauger, Linux clusters require a PhD to set up and to run. By contrast, Mac clusters are so easy to make, even 11-year-olds can do it.
"There's a book called How to Build a Better Beowulf that's 230 pages long and tells you how to set up clusters with Linux," Dauger said. "We have a one-page manual (PDF) that shows you how to do it on PowerMacs. We've had high school students do it. We've had junior high school students do it. We even had a sixth grader in Hawaii do it."
"It took NASA's Jet Propulsion Laboratory two weeks to put together a 16-node Linux cluster." he added. "I could do the same thing in less than an hour."
Dauger added that Linux clusters are extremely fragile: If all the machines in the cluster aren't running the same version of the kernel, everything grinds to a halt. By contrast, a Macintosh cluster can be made from a mix of G3 and G4 Macs running Mac OS 9 or X.
Dauger, who is 29 and a recent graduate of UCLA with a physics doctorate, formed Dauger Research a short time ago to commercialize his expertise in Macintosh cluster computing. He sells parallel-processing software, called Pooch, and offers his services as a consultant to help build clusters.
However, he hasn't done any consulting yet because all of his clients have figured it out for themselves. All they need are a few G4 Macs, some Ethernet cables, a hub and the Pooch software. Getting it up and running is as simple as installing the software and configuring it through a couple of dialog boxes.
Initially, Pooch was based on AppleScript, the scripting language built into the Mac OS, which made it easy to divide a task into subtasks and assign them to different processors.
But it wasn't very fast, so Dauger switched to Apple's networking technology Open Transport and the TCP/IP protocol. Dauger said Mac clusters have better bandwidth than similarly configured Linux clusters. They can transfer bigger chunks of data between nodes but their latency is less (The individual bytes of data are transferred less rapidly).
"It balances out," he said. "They're roughly comparable."
Dauger said Macintosh clusters, which he calls AppleSeeds, aren't yet as common as Linux ones, but systems have already been set up at schools and universities all over the world.
The AppleSeeds are mainly running scientific applications, such as physical or biological simulations, because only these kinds of applications are available at the moment. "You can duplicate a realistic simulation of Tokamak fusion generator on an 8-node PowerMac cluster," he said. "But you can't run Photoshop."
Dauger said it is easy to translate software from other parallel processors to Macintosh clusters: Applications don't even need to be modified, just recompiled.
Dauger is also hoping to convince software publishers to rewrite their software to run on Macintosh clusters. Applications such as Adobe's Photoshop, Apple's Final Cut Pro video editing package, and Alias/Wavefront's 3-D modeler Maya are all applications he'd love to see ported to parallel processors. He said tasks that currently take days or weeks could be reduced to hours on a small cluster of G4 Macs. "It's no harder to write for multiple processors than to write for two processors," Dauger said.
Mac OS X features symmetric multiprocessing, which allows the operating system and all the applications running on it to take advantage of two processors. Previous versions of the Mac OS required the software to be specially written to recognize more than one processor. Needless to say, not many Macintosh applications are dual-processor ready, and even fewer are available for more than two processors.
Pooch is a Macintosh application designed to make operating a parallel computer as easy as possible.
Pooch provides a Macintosh user interface for distributing and initiating a numerically intensive parallel application on a network of Macintoshes. It coordinates the distribution of data, acts out commands from other Pooches, and provides a user interface to start and monitor parallel computing jobs.
Pooch is also scriptable, which means you can create customized and automated queuing and launching mechanisms for parallel computing jobs using AppleScript. Because AppleScripts can be written and run from X's Unix command line, you can direct Pooch from a command line interface.
In addition, other applications can ask Pooch to query the network and submit parallel jobs, allowing automated parallel launching -- a feature of the "computational grid" -- from a Mac application. See the AltiVec Fractal Carbon demo for an example.
Requirements: Power Macintoshes connected using a TCP/IP network (100BaseT, 10BaseT, Gigabit, Airport, etc.) run Mac OS 9 and CarbonLib 1.2 or later. Pooch on OS X 10.1 and later is fully supported. (For the 10.1 release of OS X, Apple fixed many of the bugs present in previous versions.)
