To the best of our knowledge, we are the first community college to attempt this, and the first group to do so just for the hell of it, with no outside motivation to undertake this project.
We plan to keep the community updated through the net, and encourage anyone who wishes to replicate, and/or improve upon our developments to do so.
As of yet, we have received no promises as to the future of our machine. Our hopes are that it may be used in conjunction with the parallel computing class offered here, and that it would be integrated into the college network to allow entire classes access at one time. Yet, for all we know, after the current team is out of Oakton, it will end up a glorified paperweight.
** Items supplied by school, type unknown.
Sam wanted a drab, boxy, modular design, essentially nothing more than six homemade desktop boxes : two stacks three high, side by side, with a cover over the unit and a single power unit on the back.
Andy wanted to use a stretch limo with a jacuzzi. He figured by mounting the boards in the doors, and running the A/C, we could keep the system cool, and compute in style.
Eli wanted to use a two foot tall octagon, with six of the panels holding motherboards and such, and the last two panels for peripherals and monitoring the system.
Mike's idea is this, the Hyper Cube.
At this point, the design has been finalized, and construction has begun. The final design most closely resembles Eli's octagon.
Two 100MHz DX4's and a 66MHz DX2 are now running Linux 1.2.13. We have them networked with RFS and the MPI libraries installed, and have run a few parallel programs on them. Preliminary results are positive. Finding pi on two machines in parallel took about 12 CPU cycles, whereas on one machine it took about 25 cycles. Hopefully, with six machines we will be able to outpace the NeXT cluster here at our school.
We actually thought that building he chassis would be simple compared to the computing side of things. We're just dumb that way. The things that we have learned about design are invaluable. The main leson is that it's the little things that can make or break a design. Things like the mounting of hardware, the design of the panels to cover the unit, and the top of the unit have all been redesigned many times since the conception of the unit, however little conflicts and lack of thorough planning have forced most of the CAD and hand drawn designs to be tossed out the window. Most of the critical design comes from sitting in Mike's garage, arc welder in hand, brainstorming until something comes up that will work.
Even with the confusion this causes, we are almost done! Soon we'll probably have the final designs available here.
Our deadline is coming up. May 6th. Yikes!
3/24/96 - We had all the drives up and running. We were almost ready to start networking. Then it happened. As we were recompiling the kernel on one machine, the power supply started smoking and melted a bit. Whoops. Later, another power supply did the same. It turns out that our supplier did not send us UL-listed power supplies like they were supposed to, and they have been blowing up on us, so we are seeing what recourse we have towards getting new equipment. We don't know if the bad supplies have damaged the motherboards or other components, so we're going to ask the suppliers to replace everything.
On a brighter note, Mike's HyperCube has been abandoned for the Octagon. We don't have a cool name for it yet. The reasons for this include ease of assembly, space considerations, transportability, aesthetics, and ease of routing cables between machines.
3/17/96 - After finally building the interface to be able to use our scavenged monitor, we found out that we need to do a few repairs to make it work properly. The picture is a bit off kilter, so we're going to adjust the tube from the inside.
We have three of our hard drives setup. By Friday, we should be done with the rest. We were having some difficulty with LILO on the 1G drive, but those have been solved.
3/11/96 - We are currently having a bit of trouble with video cards. We've got a variety of cards ranging from VGA to high end cards, but as of yet none of them work as well as they should. Maybe it's because they have all been taken from old, broken down machines. Hopefully we will be getting a decent card soon.
We also have built and installed the LED cpu meter described in the system/Status/meter-0.2.tgz file on sunsite. It seems to work fairly well (especially the Knight Rider mode!). We are still looking at the end of Spring Break to be ready to roll.
3/7/96 - We just got our equipment in. Christmas was never this much fun. Boxes everywhere, parts all over the room. We have Linux up and running on our server, and we are working on the slave units. With luck, we will have the system ready to go by the end of Spring Break.
Our next step will be to start welding the chassis together, and installing the components.
Bob "Boom-boom" Sompolski
Joe "Cha-cha" Kotowski
Computer Builders for great service, and great prices.
Littlefuse for donating our original machines that got us started.
Mr. Wuehr for the materials to build the chassis.
