Ten years ago, Donald Becker and Thomas Sterling built a 16-node cluster, the original Beowulf, and started Linux and commodity hardware on a program of relentless improvement.
In May 1965, IBM Chairman T. J. Watson, Jr., wrote of large scientific computers, “at some point between two and three years ago it became evident that the fallout from the building of such large-scale machines was so great as to justify their continuance at almost any cost” (on Dr Mark Smotherman's site at www.cs.clemson.edu/~mark/acs.html). That doesn't mean that high-performance computing (HPC) startups have an easy time entering the enterprise market. From Control Data to Thinking Machines, history shows that if you concentrate on winning in HPC, you don't get the skills to cross over to regular business customers.
But ten years after the first Beowulf, examples of what Watson called fallout are everywhere on the Linux scene. From smoking out bad power supplies to fixing device drivers to making manageability work for the PC architecture, HPC customers are ruthless in demanding tweaks to turn a rack of off-the-shelf stuff from a maintenance nightmare to an asset. For an IT vendor, an HPC program can work like an automaker's racing program to test cutting-edge ideas and get everyone fired up to win bragging rights.
Early Linux clusters were labor-intensive, with “crash carts” including keyboard and monitor for BIOS access. Today, LinuxBIOS makes the pit crew's work feasible for more and more machines per administrator. See Bernard Li's article on how to take advantage of years of cluster experience from some of the biggest, most innovative Linux supercomputing sites (page 52).
And, forget about manageability for a while—let's talk performance issues. Paul Terry, Amar Shan and Pentti Huttunen might have just sped up many people's work by a whole lot. Check out their scheduler performance numbers on page 68 and prepare for more efficient work on your parallel jobs.
Leigh Orf has some great imagery of thunderstorms, rendered just for this issue, and the software behind them is something you can download and hack yourself. Get some ideas about scientific visualization on page 62, and send us some images.
This issue isn't all clusters—Andres Benitez and Vicente Gonzales show how they turned inexpensive non-networked air conditioners into a money-saving system for a classroom building (page 44). And Nick Moffitt, whose spam-fighting articles have been a hit on the Linux Journal Web site, is here with an introduction to a new, flexible revision control system (page 90).
Whether you're putting together a cluster or enjoying the benefits of Beowulf-driven improvements in hardware, Linux and related tools, have a great time experimenting with all the amazing technology and cool projects in this issue.