I'd like to thank all OpenSSI developers for their excellent job.
One year ago I have finished a software project
in molecular biochemistry, which runs parallel at
SMP very nice. A few days ago I have tried another SSI
clustering software before trying OpenSSI, but it was
inapplicable for my project because:
1) my processes where quite short
2) my processes where quite I/O-intensive
3) cluster just crushed very often :(
Today I decided to try OpenSSI.
Incredible, but I got 3 nodes cluster running
after *2 hours* only after getting the install
instructions. The nodes are:
- two single-CPU Pentium III 1GHz nodes and
- one dual-CPU Pentium III 1Ghz node.
All nodes of such a 4GHz experimental cluster were busy
with my first small dummy test :)
20 minutes later I was able to start real math calculation
from my project being applied to one of the shortest task.
Parallelization was very good, with 4 CPUs running the
Serial-ATA disk became busy badly. So, now I have
to think switching to RAID :)
in 3 hours I have found just a small strange thing:
CPU-bound processes of my first dummy test were
jumping from one idle CPU to another idle CPU
without any visible reason... Another thing is that
i didn't find easily the documentation on any
CPU monitoring tool, as result i have used my own "tool":
for((i=0;i<1000;i++)); do sleep 1; cat /proc/cluster/node*/load | xargs
Well, to summarize my first experience:
the first impression I got after 3 hours of using OpenSSI
is "much more then I expected", that means "very very good".
I can't say more just because of the fact that
three hours is too few for saying "excellent" :)
So, once more: my kind thanks to all cool OpenSSI developers
and keep on your great hacking!!!
P.S. I saw, the traffic is not that high yet, so I've decided this
mail should be OK.
From: Brian J. Watson <Brian.J.W<atson@hp...> - 2004-03-22 22:50:43
Khamenia, Valery wrote:
> Incredible, but I got 3 nodes cluster running
> after *2 hours* only after getting the install
> instructions. The nodes are:
Excellent. I'm glad you found it to be easy to install.
> in 3 hours I have found just a small strange thing:
> CPU-bound processes of my first dummy test were
> jumping from one idle CPU to another idle CPU
> without any visible reason...
Do you mean the processes are migrating between nodes without any
visible reason (which is controlled by OpenSSI), rather than between
CPUs on a single node (which OpenSSI does not control)?
Laura, do you have any thoughts about this?
> Another thing is that
> i didn't find easily the documentation on any
> CPU monitoring tool, as result i have used my own "tool":
> for((i=0;i<1000;i++)); do sleep 1; cat /proc/cluster/node*/load | xargs
> echo; done
A simpler (and more descriptive) version might be:
while((1)); do sleep 1; date; loads; done
The loads command displays the contents of /proc/cluster/node*/load with
an indication of which load applies to which node.
I don't think there are currently any more sophisticated CPU monitoring
tools, although I've heard of some people adapting the
demo-proclb-monitor script in http://www.openssi.org/demo-proclb.tar.bz2
to monitor process load-balancing.
> P.S. I saw, the traffic is not that high yet, so I've decided this
> mail should be OK.
Not a problem. Thanks for the feedback,