Home » Technology » Linux/Unix » Parallel processing with Unix

Parallel processing with Unix

© Robin Friedman

At the recent LinuxWorld expo in San Jose, IBM displayed a stunning demonstration of parallel processing with an off-the-shelf copy of Red Hat Linux. Parallel processing effectively combines two or more computers in to one computer, increasing its power with each added processor.

IBM broke the Cray supercomputing benchmark record with a combination of nothing greater than 300 Megahertz PCs. This astounding development means that any company or even individual needing a supercomputer can make one using only commercial products and software. The use of Linux, or another stripped-down Unix clone, is instrumental in maximizing this supercomputing power, because it requires less resources for the system than Windows and others do.

Basically, each individual computer, or node, should do nothing more than network, or communicate, with the server and follow its instructions. The server tells these nodes what to do, so the more efficient the operating system on the nodes is, the faster things run.

There are basically two approaches to parallel processing.

The first is the software approach, epitomized by PVM. PVM requires that you download free software to be able to split processes to nodes. The advantage to this is that Windows machines can be used also, including Windows 95/98 as well as NT. The disadvantage is that the configuration may take longer, and speed is slightly decreased.

The other is the Beowulf approach, which IBM used in its demonstration. This can be done with no additional software through configuration of Unix host trusting and NFS. The disadvantage is all nodes must use Unix, and configuration may be very difficult. The advantages are increased efficiency and that each node need not have a display, video card, sound card, or hard drive, in short each node may be a processor with a networking card.

This is another of Unix's prime advantages ushered into the spotlight due to Linux's explosion in popularity. Unfortunately, few companies have looked into using parallel processing supercomputers because it doesn't have an industry powerhouse backing it up. Perhaps if PVM, a government project, were taken over by a company that could gaurantee its success and support of it, parallel processing could be the next big wave in corporate computing.

IBM has proved that it can work. Keep an eye out for it, or somebody else, trying to move in on this territory.

The copyright of the article Parallel processing with Unix in Linux/Unix is owned by . Permission to republish Parallel processing with Unix in print or online must be granted by the author in writing.