Rolf Leggewie
2004-11-29 00:17:08 UTC
Hi,
just getting my feet wet with openMosix, so please be patient. I am
interested in running climateprediction.net on an openMosix cluster.
The model is huge and requires significant computing power, just perfect
for a cluster.
Unfortunately it appears that the BOINC software (boinc.berkeley.edu)
runs as a single process. I am unsure whether changes to have it run in
multiple threads would have to be done with boinc or CPDN (the
climateprediction.net client). Can anybody answer this?
Is there documentation anywhere on an openMosix friendly style of
programming (à la 'use function x instead of y')?
http://howto.x-tend.be/openMosix-HOWTO/x1661.html#AEN1665 says to just
write your code as usual and indeed that is the big advantage of
openMosix but what I'd be looking for is something that explains how to
go from single-process programs to multi-process ones.
So far it is only possible to run one workunit per node as I was told in
http://www.ssl.berkeley.edu/pipermail/boinc_projects/2004-November/000636.html.
I was unable to answer the question raised there as to whether
openMosix will appear to the software as a single CPU or SMP machine.
My guess would be that it appears as a UP machine. Will BOINC need
patches to run more than one instance of CPDN on an openMosix cluster?
Of course the optimal thing would be to run just one CPDN workunit on
the cluster and have the workload distributed over the nodes. As
mentioned on http://boinc.berkeley.edu/client_sched.php it is generally
preferable to "have one result from a project complete in time T than to
have two results simultaneously complete in time 2T".
So there do appear to be some changes necessary in the software to fully
take advantage of the openMosix cluster. Should they rather be made in
BOINC or the individual project software?
Best regards
Rolf
just getting my feet wet with openMosix, so please be patient. I am
interested in running climateprediction.net on an openMosix cluster.
The model is huge and requires significant computing power, just perfect
for a cluster.
Unfortunately it appears that the BOINC software (boinc.berkeley.edu)
runs as a single process. I am unsure whether changes to have it run in
multiple threads would have to be done with boinc or CPDN (the
climateprediction.net client). Can anybody answer this?
Is there documentation anywhere on an openMosix friendly style of
programming (à la 'use function x instead of y')?
http://howto.x-tend.be/openMosix-HOWTO/x1661.html#AEN1665 says to just
write your code as usual and indeed that is the big advantage of
openMosix but what I'd be looking for is something that explains how to
go from single-process programs to multi-process ones.
So far it is only possible to run one workunit per node as I was told in
http://www.ssl.berkeley.edu/pipermail/boinc_projects/2004-November/000636.html.
I was unable to answer the question raised there as to whether
openMosix will appear to the software as a single CPU or SMP machine.
My guess would be that it appears as a UP machine. Will BOINC need
patches to run more than one instance of CPDN on an openMosix cluster?
Of course the optimal thing would be to run just one CPDN workunit on
the cluster and have the workload distributed over the nodes. As
mentioned on http://boinc.berkeley.edu/client_sched.php it is generally
preferable to "have one result from a project complete in time T than to
have two results simultaneously complete in time 2T".
So there do appear to be some changes necessary in the software to fully
take advantage of the openMosix cluster. Should they rather be made in
BOINC or the individual project software?
Best regards
Rolf