How Do I Setup a Linux RedHat Cluster

[Mr. $T$]

I need to know how (just for fun) to set up a small linux cluster using a 10base router 2 Dual Ppros, 166 Cirix (sp?) through a 1.4 1600+ Computer.

 

[PolyPill]

Processing cluster or service clustering?

 

[Mr. $T$]

Processing cluster for Folding

 

[SickBoy]

Using a cluster for folding is pointless, IMHO. You have to run one instance per processor anyway, since the client is single threaded.

 

[Mr. $T$]

aww bummer.

Will ir run on a 166 Cirix (sp?)


And on both of the dual Ppros, I will just have to run two clients? one for each processor?

Which would be faster 2k or Redhat 7.0

 

[rogerdugans]

I have a small DISKLESS cluster running for another DC project (although I Fold as well) and while I disagree about it being pointless, it won't do what I think you want- SickBoy is right there.

The only advantage you get at all is that Diskless Cluster boxes don't need a hard drive or OS- makes them less expensive and faster to setup.
The data processing is still done by the individual machines though.

 

[PolyPill]

You can set up a beowulf cluster. I did it once a long time ago, can't remember exactly how to do it. Although you might need to do some hacking to get the folding program to run on a cluster.

This howto seams to be right.
http://www.fysik.dtu.dk/CAMP/cluster-howto.html

 

[Mr. $T$]

I guess I need to learn me some better Linux before I try that!!


Thxs guys

 

[SickBoy]

aww bummer.

Will ir run on a 166 Cirix (sp?)


And on both of the dual Ppros, I will just have to run two clients? one for each processor?

Which would be faster 2k or Redhat 7.0

Like I said earlier, you have to run one instance per processor.

I'm not totally sure if this is true anymore but I know back in the day with Folding (in the v1 days) the Linux client was weak because Linux didn't have a good FORTRAN compiler. Like I say I am not sure if that's the case anymore. If it's not, then a tweaked minimal Linux distro would most likely be faster.

 

[dustin]

Unless folding was specifically coded with pvm or mpi support even if you did have a cluster it wouldn't do what you want it to.


Here's a basic rundown of the steps to create a cluster of workstations.. I don't know much about beowulf and as far as i'm concerned there is no difference.

Install OS of choice on each box that will be in the cluster (preferably unix like OS) .. unless of course you're going the diskless client route. Do all the network/system setup gravy and add an entry for each computer in the cluster to /etc/hosts. Setup NIS+ or yp on the server node to share your essential files passwd and the like. Enable NFS daemon on server node and configure clients to mount filesystems from server (it's not considered a cluster so much if atleast one filesystem is shared among the nodes.. most used is /home or /usr/home whichever you prefer). Install your messaging api .. pvm or mpi being the most popular. There are plenty of tutorials out there on each one of these steps, it'll just take a bit of searching. Like I said, a program that is not specifically coded for a cluster is only going to run on one node, hence the need for mpi or pvm.

Have Fun