Uh, You Lose
[ The March 2005 “Booting Up”] starts, “If open source has a mantra, it must surely be, ‘Free as in speech, not free as in beer.’”
Uh, you lose. The Open Source movement began because it objected to that mantra. The Open Source movement was founded in order to avoid having to sell the term “freedom” to business people. It is not in general something they are interested in.
“Free as in speech, not free as in beer.” is the mantra of the Free Software movement, not the Open Source movement. The Open Source mantra might be something like, “Unencumbered circulation makes for efficient development,” or something like that.
David Kastrup, Bochum, Germany
Crunching the 1-2-3’s
I read the March 2005 “Extreme Linux” article about clustering PCs.
I am very interested in this concept. I do volunteer work at a local school, rebuilding PCs that are discarded to make them useful for the school. I would like to show the children how to build a “super” computer. We have eight “matched” Pentium 3 500 MHz desktops “in the shop.”
Could you recommend some sources of free cluster software and perhaps an application that would demonstrate the power of the new cluster?
John Savarese, Jr., New York, NY
“Extreme Linux” columnist Forrest Hoffman replies: You may be interested in the Stone SouperComputer,
a machine[Oak Ridge National Laboratory]built back in 1997-1998 out of surplus machines. You can read about it at http://stonesoup.esd.ornl.gov/.That page also contains links to more resources and information. Thanks for reading Linux Magazine
and “Extreme Linux”!
Apple of My Pi
I read with interest your article in Linux Magazine about a low-cost, do-it-yourself computer cluster. I did something similar with a bunch of old Macintosh desktops.
I turned five Macs (three 200 MHz CPUs, one 220 MHz CPU, and one 350 MHz CPU) into a home-grown cluster using Appletalk and two older Mac OS 9 software packages called Den Mother and Puppies. Although there was no load balancing and no way to account for processor speed, Den Mother could divide the work of a Mandelbrot computation between all five machines. If the cluster was limited to only the three identical CPUs, the computation ran in a little over the half the time as any one of 200 MHz CPUs. However, two machines ran faster than three! There appeared to be a point of diminishing return, probably due to the 10baseT connection between the machines.
But I dismantled the cluster because I csetuouldn’t find a problem to solve. I really had hoped that the Den Mother and Puppies software would allow me to run WordPerfect, Excel, and other domestic software with the aid of multiple processors, but that doesn’t seem to be the case. I’ve come to the conclusion that clustering doesn’t seem worth it for the general householder other than to be able to connect hard drives and file transfer activities which I still do.
So, what will you do with your four-CPU cluster, which by the way looked very attractive in the article picture. What problems or services do you expect to address with a home cluster?
George Dombi, via email
Forrest Hoffman replies: I set up my the home cluster just to show my readers how easy it is to do. But I actually develop parallel scientific models for my normal job, so this little system is a great place to test algorithm scalability.
The very scalability problem you encountered with your Den Mother demo is the kind of problem I look for when developing applications. Much of the overhead you encounter with communications over Ethernet
is the large latency associated with opening the connection through the kernel. As a result, hardware vendors have developed low-latency, high-bandwidth interconnects to solve just that kind of problem. I have written a piece about such interconnects. See http://www.linux-mag.com/extreme/.
