I am surprised FAH doesn't have clusters

[Shadowmeph]

I just learn a tiny bit about Clusters today ( hak5 website) and so I did a bit of a search on the net, I am surprised FAH doesn't have clusters I mean Clusters that would work on one unit at a time. there most be some one out there that can write a script for this, the closest thing I found was Pelicanhtc PelicanHPC: A GNU/Linux distribution to create a HPC cluster for MPI based parallel computing

[ilya]

The F@H project is a computing cluster in itself. Pande explained your question somewhere on his blog IIRC.

[Shadowmeph]

Originally Posted by ilya

The F@H project is a computing cluster in itself. Pande explained your question somewhere on his blog IIRC.

I am pretty new to folding so where is the blog you mention.

[Dead Things]

Folding@home

ilya is correct - fah is itself one giant cluster, and each client is a single node, hence the inability to scale across multiple nodes. Note - the GROMACS core does in fact support clusters, but the fah implementation of it does not owing to the method it uses for assigning work to nodes.

Look at awachs, for example - he's got a powerful supercomputer there, but instead of running one humungous bigadv client, he runs a myriad of regular SMP clients because he can not scale across multiples of his own nodes.

[Shadowmeph]

I guess that it is a cluster in a sense but each node is independent or maybe a better phrase would be " works independently from the main until completed".

I have laying around my place a q6600 on a dfi p35 lan party and a AMD 955be on a 790 board and hopefully very soon I will have my AMD 6128 CPU s so I can fire up that. is it possible for me to get all of those up and running in some sort of cluster to work on humongous bigadv client , or would I have to run 3 separate ones. I currently am running my AMD 1100 and my q6600 quad separately . I would prefer to not use my main pc to fold at all

[Shadowmeph]

ok I just thought about all of this an I realized that it would be way to hard to setup all machines to work on one thing because of the speeds of the different machines alone would mess things up.

[Dead Things]

Originally Posted by Shadowmeph

ok I just thought about all of this an I realized that it would be impossible to setup all machines to work on one thing because of the speeds of the different machines alone would mess things up.

FTFY

Just think about what SMP stands for and you'll realize why clustering single work units cannot be done. Clusters work when the process requires intermittent coordination between nodes - usually at the beginning and end of specific sub-tasks. SMP processes, meanwhile, depend upon constant coordination between nodes.