Distributed Computing Within Distributed Computing

[Wolfmist]

Wouldn't it be nice to have a feature in the FAH client to use a bunch of computers in a network all going at it collectively for one individual unit?

As the title says, distributed computing within distributed computing. This would be nice for when you have a limited amount of time with a large amount of computers.

BTW way, in terms of weekly production I am around the 30th top producer for team 32 at around 90-100 pts a week. If anyone would like feed my addiction by surrendering their FAH names and replacing it with Wolmist, let me know ;)

Also I have been able to run the text FAH client as a service using firedaemon (www.firedaemon.com). It has worked flawlessly so far. If anyone needs any questions about that let me know.

[KFB]

i think that is like running F@H on a Beowulf cluster. if the F@H client were multi-threaded then maybe the cluster's nodes' work would be aggregate.

[Wolfmist]

I've read the term beowulf cluster before. Could you explain it a little more?

[KFB]

ok, you asked....
Cut & paste from http://www.beowulf.org:


The Beowulf Project was started by Donald Becker when he moved to CESDIS in early 1994. CESDIS was located at NASA's Goddard Space Flight Center, and was operated for NASA by USRA.
In the summer of 1994 the first Beowulf 16 node demonstration cluster was constructed for the Earth and Space Sciences project, (ESS). The project quickly spread to other NASA sites, other R&D labs and to universities around the world. The project's scope and the number of Beowulf installations have grown over the years
The Beowulf Project is now hosted by Scyld Computing Corporation, which was founded by members of the original Beowulf team with a mission to develop and support Beowulf systems in the larger commercial arena.


visit the beowulf.org site and follow links...
a few more links on clustering:
www.aggregate.org (http://www.aggregate.org/)
Building Linux Beowulf Clusters (http://fscked.org/writings/clusters/cluster.html)

very cool stuff
:)

[Cluster]

This is my summer project. 12x1.2GHz Duron Beowulf Cluster. I think if you run multiple clients on a cluster it will run much like a dually running 2 clients. I'm gonna try it out, if not, they'll just run idividually and the beowulf client will be scrapped.

[KFB]

Originally posted by Cluster
This is my summer project. 12x1.2GHz Duron Beowulf Cluster. I think if you run multiple clients on a cluster it will run much like a dually running 2 clients. I'm gonna try it out, if not, they'll just run idividually and the beowulf client will be scrapped.

very cool summer project.
who is funding those dozen machines? do you need some help?:D

[Ploaf]

It shouldn't be too hard. The on ething that you will need is the source code and a compiler. Everything needs to be recompiled to work on your beowulf. If you can get the compiler and the source then you should be fine. the problem is getting the source code.

[walaka7]

pk that was way over my head !!!!!!!! :eek: Time to hit the books again

[KFB]

Originally posted by Ploaf
It shouldn't be too hard.
.... the problem is getting the source code.

you can say that again. i sure wish they would release the F@H code though. then i would actually have an interest in some code.

[Arkaine23]

Anyone try asking Vijay about F@H and Beowulf?

[Wolfmist]

From a quick search it seems Vijay is someone that has to do with Stanford. Since I don't know him, or where to find him, I can't ask him (maybe her?). Who is he out of curiousity.

Here's a poll:

How many of you folding freaks out there (like me) if it were implemented, would combine a bunch of computers on a network and have them all fold on a single unit?

I would.

As I think about it more, I wonder what would be the best way to do it would be. I think it would work best if it stayed like the big project. Meaning that a host computer would give out a frame to do on each of the computers.

I don't think having them all work together alternating very small parts (like dual chips) would work out because of latency and it would use too much bandwidth on a network, and it would make something like using a dial up with the group much harder.

So what do you guys think?

[res0r9lm]

I think 12 althlons folding one wu would be slower than 12 athlons folding 12 seprate wu's

[Wolfmist]

Originally posted by res0r9lm
I think 12 althlons folding one wu would be slower than 12 athlons folding 12 seprate wu's

Yeah, I'm sure they would but that wasn't my point. My first post says it all.

What I wanted was to able to use a bunch of computers for a limited amount of time. I'm a lab assistant to a college, and I can't have folding@home running while I'm not here. But if I could use the time I am here then I could knock off a quick WU, then quit F@H when I leave.

[KFB]

this discussion of parallelism is one of the classic challenges of computing. the programming involved in implementing a multi-threaded protein folding application is very complex, to say the least. IBM's Blue Gene computer, which will be capable of 200 TeraFLOPS when it is finally completed, is scheduled to run protein folding on its many processors.

[Wolfmist]

So what would that be, a WU every second?

[Ploaf]

Blue Gene/L isn't even the fastest thing they are working on. The "L" may as well mean Lite. It's part of the Blue Gene project in which they are attempting to build a petaflop scale machine. I believe that this is because they will have problems finding someone to buy the original Blue Gene so they are making a consumer version that is much smaller. ;) It will be very interesting to see what these new supercomputers are really capable of.

[silversinkbam]

Originally posted by KFB
i think that is like running F@H on a Beowulf cluster. if the F@H client were multi-threaded then maybe the cluster's nodes' work would be aggregate.

this thread lost me from this point on..........:eh?: :confused: :confused:

LMAO........notice it is first reply

[KFB]

Originally posted by Ploaf
Blue Gene/L isn't even the fastest thing they are working on. The "L" may as well mean Lite. It's part of the Blue Gene project in which they are attempting to build a petaflop scale machine. I believe that this is because they will have problems finding someone to buy the original Blue Gene so they are making a consumer version that is much smaller. ;) It will be very interesting to see what these new supercomputers are really capable of.

yes, thanks Ploaf. i thought i remembered reading that the Blue Gene project was expected to be the first petaFLOP computer, but i visited to the ibm website and saw the 200 teraFLOp figure, so i changed it before my post was submitted. hell, even the Lite version is still a very capable supercomputer. but the "Big Momma" Blue Gene will be a real milestone in computing. IBM will be responsible for yet another breakthrough.