Looking for a Fast CPU

A research paper that is being published by myself and a professor I'm working with is due on Saturday.

We're trying to add a final graph to the paper that uses data from simulations I coded (in Java). The simulations itself concerns the manipulation of permutations (mixed up numbers). The problem is that at the size we need these simulations ran (permutations of n = 1 000 000), the time it takes to complete on my computer is just enormous

I'm running a Duron 950 OC'ed to 1140 (394 mb PC133), and I've been running it for several hours now and it has yet to complete 1 trial (we need to run many trials to form this graph)

I'm wondering if someone here is able to lend me their CPU cycles (perhaps during the night or while you're away).

All you need to do is run the java program. I think that if I've been running it for 4-5 hours on my computer and I get nothing for this size (1 million permutation), then I'll have to turn to those with Pentium 4's (3 gigs? How fast will that be compared to a Duron 950?)

I know its a far cry, but if anybody would be willing to help...

You probally will need to take a distributed approach to this. Can you run a subset of permutations and combine them later?

I know when professors and students at my University needed a lot of processing they would remotly log into a bunch of Unix machines on campus.

That said, I'd be willing to run it on my 3.0c P4 with 1 GB Ram provided you give me the source code and instructions for compiling it. I would have to review it before I ran it. I wouldn't be able to start until after work so it wouldn't get started until late today at the earliest.

send it to me chum, ICQ 16858094 MSN derusett @ gametoast.com

my machine isn't MUCH faster its overclocked to 2.1Ghz, from a P4 1.6a but I'm sure I can churn them out faster then you

mccoyn said:

You probally will need to take a distributed approach to this. Can you run a subset of permutations and combine them later?

Click to expand...

Yes. Think of it like this. I have 2 parameters, k and d.

k takes on values (I haven't decided what yet, but for now...) 2000 - 20 000 with steps of 2000. This will be on the horizontal axis of my graph.

Now, I have to run the simulation a certain number of times (and take an average) at each value of k (I did 50 before, but that may be too long at the moment)

d is either 'yes' or 'no' representing two curves.

I can break this up so people can run through values of k, and I'll take the averages muself. Either way, it's hard to say at the moment, I need to test this on someone with a fast comp to see what I can and what I can't do.

Here is the program. Now the parameters can change (I can change the number of loops of a certain for statement so the parameterk will change automatically) but I set it at k = 2000 for the moment with ONE loop so I see how fast some people can crank this out.

The program you have to run is titled Amalgamater. It uses a bunch of other classes (especially Permutation). Simply run the program and enter the first menu item (type "1" then ENTER)

You can grab a zip file of all the stuff here (zip into an empty directory. The src files are in their own directory)
ZIP FILE

Or if you want a jar file of the src files:
JAR FILE

deRusett, sorry bud, didn't see your reply when I posted. Give it a try.

I really just need a test value to gauge how I'll do this in the next day or two.

Oh yeah, if it's taking too long (you be the judge, I ran it for 5 hours),. then don't bother (it'll probably end up too long in the end anyways without changing parameters since I need to average several values)

I don't want anybody troubled

is it not compiled? I can't get it to run under XP, all I have are Class files, and your src/pack/Java files all open as text. I've never used Java files out side of Unix

You might be able to find lots of volunteers with lots of Ghz in the Folding Team forum.

I don't know how complicated the program is, but if it's short I would consider rewriting in C++ or mathematica/matlab custom programming languages. Java is extremely slow for that kind of thing, probably your worst choice of language, as it's not really compiled. You would have orders of magnitude more speed with a compiled language, and even more if you wrote it in assembler, although that's a serious pain in the ***.

If you can get a good distributed computing setup going, I can run some for you. I have an Athlon 3000+, Athlon 2100+, Athlon 1.4 Tbird, and K6-2+ 550. Let me know if your trials show that it's possible to get it done.

Also, post in the folding forums, you'll find a lot of distributed computing people with more computing power than they know what to do with there. =p

Sorry guys, hang tight, I'm waiting on some stuff.

There is also a portion of my code that I may be thinking of optimizing (and if I'm correct, this will literally shave off hours)

I had a friend with a pentium 2.6 Ghz run it. It took him something like 10 hours, so right now, it's much too long for any of you

hehe I was only Vollenteering because I am taking off for 3 days and have 2 Machines that can do it for you while I'm gone.

Have you tried running this thing through a profiler? I believe there are several open source java profilers available, try searching www.sourceforge.net for one that works on your platform. It's best to know where the speed problem actually lies, so you can optimize only that part, rather than trying to guess at it.