Welcome

Results, conclusions

After connecting everything, I decided to compile a kernel. I took the 2.6.17.13-kernel, unpacked it at my workstation, entered the options to make it compile on my 10 terminals, and timed the compile. It took 22 minutes. That was a tad of a dissapointment: the same compile took 6 minutes on my 2000MHz AMD64-3000. While the 10 processors are 200MHz each and the total amount of GHz should be the same, the overhead of the network and OS, combined with the fact that the Geode GX1-architecture of the terminals is more than a few years old, made for a cluster that is 1/4 as fast as my main machine.

People tend to ask me the question why I hack stuff. Sometimes it gives results one can't get by buying off-the-shelf stuff, but most of the time my time could be better invested by going to work and use the extra money to buy a device that can do the same. The reason I do it is not because I don't know that, but because I like doing it, it entertains and educates me. Same with this project: while it can be viewed as a failure (a second hand P3-700 doesn't cost a thing these days and has better specs than my cluster), the way to get there was interesting: I learned stuff about PCI, BIOSses, clustering, etc, and generally I enjoyed doing it. Secondly: while the end-result isn't that interesting, I can boast I'm the first one that has run Linux on these devices, and they do make great small passive little networked units.

« Prev 5 

26 comments

travis wrote at 9 Aug 2014, 16.10:

"...Same with this project: while it can be viewed as a failure...." the fact that you could possibly consider this a failure is extremely comedic. your projects consistently BLOW MY MIND... nothing is a failure if you learn something from it... and this is far from failing its super amazing... what does the cluster do now? insert Beowulf Cluster from 1995 joke here....

AndrewZ wrote at 1 Mar 2014, 15.53:

Love reading about your hacking adventures! keep up the good work!

Rena wrote at 4 Aug 2013, 5.10:

Why? Because I can, dammit.

Hardcoresmurf wrote at 10 Jul 2011, 17.05:

Mooi werk sprite_tm leuk clustertje KUTGW!!!!

Sprite-tm wrote at 1 May 2010, 16.33:

Nope, that was one kernel with the workload divided over all the terminals. That's why I was disappointed in the slowness of it all.

Jeremy wrote at 1 May 2010, 16.24:

Did it have to compile on each terminal or were you dividing the work? If they each compiled the whole thing that would more reasonable for the time.

Elmars wrote at 12 Jan 2008, 21.55:

Two suggestions... 1) those geodes can run at 333mhz. should be a set of three jumpers on the board for this. 2) I suspect the machines are a bit ram limited if you are doing a network boot. Add more ram to the machines and then try to run the cluster again.

dilan wrote at 19 Jul 2007, 15.13:

Great work. Best wishes for further Linux Works!!!!!!!!!!!!!!!!

Christian wrote at 19 May 2007, 20.52:

Hey, I'm not the only one trying weird stuff :-D

nisse wrote at 29 Apr 2007, 12.02:

Hi! Very nice reading. I love those experiments with old hardware. Keep up the good work! Best regards nisse, Stockholm, Sweden

Da JoJo wrote at 16 Apr 2007, 15.11:

So nice to know that there are more people interested in making old stuff usefull again. I liked reading your article a lot. Keep up the good works :-)

Sprite_tm wrote at 13 Mar 2007, 14.29:

Mail me and I'll send it to you.

Max wrote at 12 Mar 2007, 20.39:

hallo i will build the WT 300 as a "PC" where I can find the ROM image for WT300, I found nothing

nicOS wrote at 15 Dec 2006, 22.37:

For several years motherboard have build-in EPROM programmers. Was hotflashing not an option ? http://www.onlamp.com/pub/a/onlamp/2004/03/11/bios_hotflash.html Hotflashing means programming a BIOS chip in a different computer than it belongs (we'll call this other computer and BIOS the host). This includes removing the current BIOS of the host computer -- after booting -- and putting on the problematic one and flashing it.

Artur wrote at 10 Oct 2006, 20.42:

Great work, You are very talented and inventive person, and I realy enjoy reading about all You new projects. If You only have time keep on doing it please :)

evandrofisico wrote at 8 Oct 2006, 19.30:

I guess that the best use for these machines after being reflashed would (logically) become LTSP enabled clients. Ok, it's just a migration, but as LTSP is much more flexible, it's a great gain! and overall, your project is very interesting. Congratulations.

gotrootdude wrote at 8 Oct 2006, 16.15:

Those thin clients would make good asterisk pbx boxes. Just keep them booting from CF.

spaceride wrote at 2 Oct 2006, 7.39:

Doesn't matter it fails, great project. I wanted to build one myself, in that time I've had opportunity to buy 30 P1-P2 box (complett) for a bout 70$ with shipment, but my parents would thrown me out from home if I bring that garbage here :) So actually I did some research with 2 machines lately (AMD 2200 X 2) and windows mmpi + linux, but I was noob to that clustering topic and I haven't got too much time. So if you could give me links about clustering in linux, howtos, tutorials, parallel programming, that would be nice!

hrf3420 wrote at 2 Oct 2006, 6.11:

Hey, what about the WYSE thin clients? I have a few of those and am to put linux on them. Any ideas?

jeremy wrote at 1 Oct 2006, 22.08:

well try overclocking the core's ? if its posible

strider_mt2k wrote at 1 Oct 2006, 17.45:

Very interesting read. Sorry your results weren't what you were hoping for, but all-in all it's still great work!

MRE wrote at 1 Oct 2006, 4.17:

Well, now that linux is on it, and you can run a drive, it would make a great special purpose computer IE data collector, robot controller, Whatevermachine. Good job! Still usefull in the end.

Marco wrote at 1 Oct 2006, 1.53:

Hey, great work, I'm trying to do something like that on a IBM Netvista Thinclient. That should be much easier, because it has a inbuild CF-Reader. But I like your stuff. :)

Sprite_tm wrote at 30 Sep 2006, 21.19:

I don't think so: the fsb/processorspeed-factor is better in the terminals than in my own server (the terminals seem to run at 66MHz FSB). I heard the Geodes basically are a souped-up 486-processor with most of the accents placed on 16-bit processing without too much thought given to the 32-bit stuff; that would explain why they suck under Linux, performance-wise.

Gary wrote at 30 Sep 2006, 21.14:

Could your processing speed be a FSB bottleneck?

tyrelvnne wrote at 30 Sep 2006, 20.42:

great project really intresting. keep up the good work man..

Leave a comment:

Your name:

What does this picture say?
Sorry, this is a captcha

Your comment:


© 2006-2016 Sprite_tm - Contact