BOINC on Linux Cluster

Questions and Answers : Getting started : BOINC on Linux Cluster
Message board moderation

To post messages, you must log in.

AuthorMessage
Jacek Kasperczyk

Send message
Joined: 16 Sep 02
Posts: 4
Credit: 129
RAC: 0
Poland
Message 3052 - Posted: 1 Jul 2004, 8:38:13 UTC

Apart from the problems with getting WUs I have some major problem with running BOINC on 4-node Linux cluster (based on Mosix kernel patch). Initialy I've created 4 separate folders (named: cpu1, cpu2...) with fresh BOINC apps. Running all 4 processes leads to its migration to all nodes of my cluster (it's quite normal), but in my BOINC stats there is the only one idetified host (node1, where all 4 processes were started). I guess that it may be a problem in matter of assigning WUs, because BOINC assumes that there is only one CPU loaded with 4 BOINC processes. Any ideas?

Kind regards
Jacek
jacek@wsaiz.pl
ID: 3052 · Report as offensive
John McLeod VII
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 15 Jul 99
Posts: 24806
Credit: 790,712
RAC: 0
United States
Message 3151 - Posted: 1 Jul 2004, 14:04:58 UTC

Is this 4 CPUs in one computer? Or is is 4 computers that the OS is running as one computer with 4 CPUs? Or is it 4 separate computers? If it is the first, run one copy of BOINC, and let it run separate processes th crunch WUs on each processor. If it is the second, you may be able to run one copy of BOINC, and let BOINC start one crunch process per CPU. If it is the latter, I really don't understand what is going on.

ID: 3151 · Report as offensive
Jacek Kasperczyk

Send message
Joined: 16 Sep 02
Posts: 4
Credit: 129
RAC: 0
Poland
Message 3185 - Posted: 1 Jul 2004, 15:01:54 UTC - in response to Message 3151.  
Last modified: 1 Jul 2004, 15:05:35 UTC

ID: 3185 · Report as offensive
Jacek Kasperczyk

Send message
Joined: 16 Sep 02
Posts: 4
Credit: 129
RAC: 0
Poland
Message 3186 - Posted: 1 Jul 2004, 15:02:04 UTC - in response to Message 3151.  

There are 4 separate computers (each one has 1 CPU) clustered under Linux system with Mosix kernel patch (www.mosix.org). So it's your second guess, but i think that it's impossible for BOINC to deploy processes for each CPU in cluster. BOINC can run on SMP machines and one can define number of CPUs. Mosix cluster is a kind of MMP architecture and from logical point of view the all nodes are separate machines, where the processes can migrate to. If I set number of CPUs to 4 BOINC will try to allocate these CPUs on the machine where it was started and will fail. The point is to deploy automaticaly as many computing proceses as the number of cluster nodes is. Fortunately the project is still under heavy modifications, so I have some time to think it over.

Regards

Jacek
jacek@wsaiz.pl
ID: 3186 · Report as offensive
Viek

Send message
Joined: 3 Oct 05
Posts: 2
Credit: 32
RAC: 0
Indonesia
Message 228334 - Posted: 9 Jan 2006, 6:16:21 UTC

Hi,
Dear Jacek i've read your post,

firstly, could i ask you something about your thread on running BOINC on 4 node linux cluster
does mosix kernel patch require one working directory on NFS ?, so the slave can ssh to the master so it can work in a parallel way?

secondly, how could i run BOINC client ?, should i install it to the working directory ? or is there any other way ?

FYI: im using parallelknoppix
http://pareto.uab.es/mcreel/ParallelKnoppix/

thank you
ID: 228334 · Report as offensive
CPU BorgMeister

Send message
Joined: 16 Dec 05
Posts: 7
Credit: 434,926
RAC: 0
United States
Message 229057 - Posted: 10 Jan 2006, 23:00:39 UTC - in response to Message 228334.  

Hi,
Dear Jacek i've read your post,

firstly, could i ask you something about your thread on running BOINC on 4 node linux cluster
does mosix kernel patch require one working directory on NFS ?, so the slave can ssh to the master so it can work in a parallel way?

secondly, how could i run BOINC client ?, should i install it to the working directory ? or is there any other way ?

FYI: im using parallelknoppix
http://pareto.uab.es/mcreel/ParallelKnoppix/

thank you


I have tried makign BOINC run with both ParallelKnoppix and ClusterKnoppix. From what I can see, the current implementation of BOINC won't migrate to any other hosts. Possibly because it's implementing shared memory?

For my Mosix config, I have 40 nodes. I'm working on creating a BOINC directory and just using the MFS and DSH to manually kick off a BOINC process for each node. (If I can get the nodes to quit complaining about error while loading shared libraries: libpthread.so.0:)
ID: 229057 · Report as offensive

Questions and Answers : Getting started : BOINC on Linux Cluster


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.