question for a frenid

[crash893]

A freind of mine posted this on another board

anyidea if he could run prime95?

I know that prime95 isnt very good for multi machine setups but i dont know if it being a beowolf cluster would change that or not?

Hey all,

I work in the IT dept at a major university. One day, a post-doc came to the IT dept with a request for some computing power to do bioinformatics research. We reciently replaced a computer lab consisting of 27 P3-700 with 256MB ram each with 27 Dell GX280's with 1GB ram each, so we had a ton of the p3's laying around. An important note is that the p3 machines were all built by the it dept on tyan smp mobo's. To solve the post-doc's computing power problem, we made a beowolf cluster consisting of 7 dual p3-700's with 512MB ram each and a head node of a dual p3-700 with 1.5GB ram (I think that is the final configuration). After a weekend of crunching 1000's of jobs, the cluster now sits there doing nothing.

I was wondering what could be done with the cluster that would be fun and interesting. It is currently running Rocks Cluster rolled out with Red Hat 4.

Any Ideas? I could try folding@home but I don't want the university to get mad due to the amount of bandwidth used....also only the head node is connected to the university network through a second nic. the other nic connects to a switch that is connected to the compute nodes. Perhaps something that takes advantage of the Sun Grid Engine?

[Wacky]

Perhaps he could join the NFSNet effort. The Cunningham (Factoring) Project is well respected in academic circles and we can always use more help.

[Prime95]

Originally Posted by Wacky

Perhaps he could join the NFSNet effort.

I agree. P3s are a little dated for Mersenne prime hunting. They would be good for NFSNet or ECMNet.

[garo]

Or Lone Mersenne Hunters

[olmari]

Originally Posted by Prime95

I agree. P3s are a little dated for Mersenne prime hunting. They would be good for NFSNet or ECMNet.

Well, surely they are slow side, but then again it is faster than without that machine, aint it?

[Prime95]

Originally Posted by olmari

Well, surely they are slow side, but then again it is faster than without that machine, aint it?

Yes, and if you decide to search for Mersenne primes your contribution will be appreciated.

The suggestions above were made because of the P4's marked improvement in FPU performance compared to the P3. Whereas a 2.0 GHz P4 will be 6 times faster looking for Mersenne primes, it will be only 3 times faster at NFSNet or ECMNet. Therefore, your machines would make a bigger impact on those projects. Do whatever project you think is most worthwhile and fun.

Back to your original question. I think that no matter which project you run, the lack of Internet connection at each node will force you into some kind of manual assigning of work to each node.

[olmari]

Originally Posted by Prime95

Yes, and if you decide to search for Mersenne primes your contribution will be appreciated.

The suggestions above were made because of the P4's marked improvement in FPU performance compared to the P3. Whereas a 2.0 GHz P4 will be 6 times faster looking for Mersenne primes, it will be only 3 times faster at NFSNet or ECMNet. Therefore, your machines would make a bigger impact on those projects. Do whatever project you think is most worthwhile and fun.

Good point, I didn't think it that far.

[Wacky]

Originally Posted by olmari

Well, surely they are slow side, but then again it is faster than without that machine, aint it?

Not necessarily.
If a potential participant is unable/unwilling to make an adequate contribution, the overhead for the project to support the participant may exceed the value of the contribution.
Additionally, there is the serialization question. In some respects, all projects have goals that require that each part of the process be finished before the goal is reached. If one part is held up because the contributor is "too slow", the entire project may be hurt by its inability to reach the next goal.

[R.D. Silverman]

Originally Posted by Prime95

Yes, and if you decide to search for Mersenne primes your contribution will be appreciated.

The suggestions above were made because of the P4's marked improvement in FPU performance compared to the P3. Whereas a 2.0 GHz P4 will be 6 times faster looking for Mersenne primes, it will be only 3 times faster at NFSNet or ECMNet. Therefore, your machines would make a bigger impact on those projects. Do whatever project you think is most worthwhile and fun.

Back to your original question. I think that no matter which project you run, the lack of Internet connection at each node will force you into some kind of manual assigning of work to each node.

Hi,

If you would like to run NFS *offline*, I am willing to provide source,
binaries, and input data. You can send the output files back by email....

Bob

[wblipp]

Originally Posted by Prime95

I think that no matter which project you run, the lack of Internet connection at each node will force you into some kind of manual assigning of work to each node.

The ECMNET projects have a natural method to handle this communication problem. You could set up a slave server on the node with internet access and set up the all nodes with clients pointing to the slave server. With the new ECMNET version you could set up multiple slave servers, one for each project, and could have the clients time share among the projects.

The biggest disappointment in this approach is that it doesn't use the cluster aspect at all. But so far nobody has proposed anything that takes advantage of the cluster aspect.

If you go this approach, I hope that one of the servers you slave will be an oddperfect.org server. There is a "Most Wanted" server at oddperfect.no-ip.org:8201, but the large ECM level (presently 110M) is a problem for some machines, so there is also a "More Wanted" server at :8202 that hands out smaller assignments - presently 1M.

William

[crash893]

.....also only the head node is connected to the university network through a second nic.

ill pass this along when i get to work