BOINC Client and Rocks Cluster LINUX

Message boards : Questions and problems : BOINC Client and Rocks Cluster LINUX
Message board moderation

To post messages, you must log in.

AuthorMessage
Jeremy

Send message
Joined: 20 Jul 08
Posts: 1
United States
Message 18673 - Posted: 20 Jul 2008, 19:02:37 UTC

While I consider myself fairly computer/network competent, I would appreciate any assistance in the following setup.

Here is my plan:

Use Rocks open-source Linux cluster (http://www.rocksclusters.org/) to connect the following 12 workstations:

Dual Intel Xeon 2.4GHz Processors with 1.0GB ECC RAM, Two 36GB U320 SCSI Drives and a 10/100/1000 NIC.

Has anyone tried a similar cluster with BOINC?

Do I need to install the BOINC client on the Frontend Node as well as all the Compute nodes?

How do I get everything to work together?

Will there be any difference on running these dozen computers as a cluster vs. 12 individual workstations?

Any help on this matter would be greatly appreciated.
ID: 18673 · Report as offensive
Les Bayliss
Help desk expert

Send message
Joined: 25 Nov 05
Posts: 1654
Australia
Message 18702 - Posted: 21 Jul 2008, 1:28:26 UTC

The easiest way for you to get a reply from someone who knows the inner workings of BOINC is to post my idea of how it works, and then wait for someone to correct me.

So,
BOINC was designed back in the days of single core computers, for people with single computers, in the hope that enough people individually running a small part of a problem, in isolation from each other, would allow a project to collect the small results and use them in the manner of jigsaw pieces to build up a larger picture.
This still applies today, even for multi-core computers, and people with large collections of computers; each core of each computer works in isolation to run/crunch a small part of a problem, with the project putting the pieces together.

If you have/are going to create a cluster computer, you'll still only be able to run a copy of BOINC on each machine, with each core running a separate work unit in isolation from the others.

And keep in mind that BOINC doesn't do any of the crunching; it's the science applications that do that, and they've been designed to run in isolation, and not in parallel.

ID: 18702 · Report as offensive
Eric Myers
Avatar

Send message
Joined: 12 Feb 06
Posts: 232
United States
Message 18723 - Posted: 22 Jul 2008, 3:09:41 UTC

You'd likely want a squid proxy on the head node, so that it can forward connections from the nodes to the project server.
-- Eric Myers

"Education is not the filling of a pail, but the lighting of a fire." -- William Butler Yeats
ID: 18723 · Report as offensive
John McLeod VII
Avatar

Send message
Joined: 29 Aug 05
Posts: 147
Message 18804 - Posted: 24 Jul 2008, 12:55:57 UTC

BOINC only works in homogeneous systems. All the CPUs have to be the same. So, in your case, install BOINC on all of the computers that you want BOINC to be running. Either allow the compute nodes out to the internet directly or install a proxy/router in the head node. Each node would be completely optional. You can have some nodes with BOINC on them and others without, head node included.

BOINC WIKI
ID: 18804 · Report as offensive

Message boards : Questions and problems : BOINC Client and Rocks Cluster LINUX

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.