Parallel Computing With Ordinary Computers

[Vorticity]

I've been thinking about this for a while.

As a geek, I have accumulated quite a few old computers over the years that languish in a closet. Does anyone know if it is possible to set up a parallel computing structure by hooking up several ordinary PC's?

If so, what kind of software would be needed and, more importantly, how would the computers be linked? Direct Cable Connection (too slow)? Some kind of network thingy?

Has anyone done this before? I've googled, but can find nothing.

I think I can imagine ways of slapping together some bare-bones code that could accomplish this rather poorly, but I'm wondering if some uber-geek somewhere has figured out how to do it in a more nifty way.

Thanks.

[scribble]

Originally posted by Vorticity
Thanks. [/B]

http://www.beowulf.org/

"Imagine a beowulf cluster of those!?"

http://www.phy.duke.edu/~rgb/Beowulf/beowulf_book/beowulf_book/index.html

[a_unique_person]

Of course it can be done, but parallel computing is a quantum leap over the traditional model of computing. One of the things that has always held it up is Moores Law. For the average user, there has never been a compelling reason to move to multiprocessing because the user PC is getting faster and cheaper quicker than the reason to use multiple computers.

For the same reason, if you have a Pentium I to hook up because it is lying around doing nothing, it isn't going to add much to your computing power.

Mainframe computers are moving forward. I used to work with IBM mainframes, but the average Cobol hacker wasn't doing much to make use of the power available. Hell, for years CICS, the IBM Transaction Processing System, wasn't able to use more than one CPU at a time, and it was supporting thousands of users at once.

They can now support multiple operating systems simultaneously, and users, and CPUs. If a CPU suffers a hardware failure, it will move the program to another CPU, and carry on like nothing happened.

Yet at the same time, there are computers out there that support multiple CPUs, eg, weather forecasting systems, but these are specifically written to use the CPUs.

An operating system that just hands out tasks to CPUs doesn't exist, as far as I know. BEOS was in existence for a while, but AFIK, it was able to use multiple CPUs when the software was written to take advantage of them. Once again, the automatic sharing of work was not there.

Given the amount of work that has gone into even getting to where computing is at these days, I would guess that is because we just aren't smart enough to do it yet commercially and programatically.


Of course, I could be wrong.

[Wudang]

How about grid computing? (http://www-130.ibm.com/developerworks/grid)

[iain]

Not only is it possible, it's been happening for years. SETI is an example of a distributed parallel computing network.

The main feature of normal parallel computing is that a central system can break down a workload into smaller packages and parcel them out to the nodes of the parallel computer (which could be normal PCs). They are ideal for big batch jobs; far less appropriate for other computing tasks.

In terms of operating systems, Beowolf on Linux does the job, as does PSSP (Parallel Systems Support Programs) on AIX (the OS the computers like Deep Thought run on - they were parallel computers).

If you look at supercomputers, they are all parallel computers with hundreds or thousands of nodes.

As AUP says, though, a slow computer is still a slow computer no matter how you cut it.