Re: CCL:Origin300, Linux Cluster
Manuel Melle-Franco wrote:
I can only support this.
And I would also suggest using no no-name, garage-built hardware.
Under which basis? If you want to get the best prize/performance
this is the best choice, full stop!. You got to study in deep all
possibilities, but regardless...
Sure. The basis for my suggestion is my(!) best trade-off between the
amount of non-scientific work that I have to invest and what I finally
get out of it scientifically.
If you run into problems with your garage-built hardware (which can
without any doubt be as good as any other hardware) you might be stuck
with it and you might have to fix the stuff yourself.
A major brand vendor producing server hardware can just not afford
building sh*t because otherwise they are out of business within weeks
because the 24 h exchange support eats all there profits.
Since I could not afford being without my compute server for more than
couple of days, we decided to go for a brand-name hardware.
All major rack-server vendors, including ours, come and exchange your
hardware within 24h in case of a defect. I just had this case a couple
of days ago (processor heat death due to a damage fan) and I was really
happy to see my server running again after 24 h without having to do
more than a phone call.
Just checked and best you can get for a reasonable prize there seems dual
PIII. That was OK maybe last year, but now for not a much higher price,
but no big brand (the ones I know, Siemens, compaq, dell, and IBM, do not
sell AMD's and dual P4 are still too expensive). You can get dual ADM
athlon 2000 which definitively runs faster for mainly the same prize but
you have to renounce to the big brand name.
We are running Siemens hardware and we are very
satisfied, but I think
there are tons of comparable systems from other major producers.
I am pretty sure, I am also pretty sure that they are not the best
alternative nowadays, :-)
There are always systems with newer, faster, etc., hardware. The
I'm talking about are designed for stability, which might cause the
vendors to stay with a particular hardware combination for a few month
longer. We have had half a year of uptime with some of these and the
reboot that interupted this was more like a vodoo idea ("hm, it
good to reboot from time to time, no one knows...").
Dr. Christoph Steinbeck (http://www.ice.mpg.de/departments/ChemInf)
MPI of Chemical Ecology, Winzerlaer Str. 10, Beutenberg Campus, 07745
Tel: +49(0)3641 571263 - Fax: +49(0)3641 571202
What is man but that lofty spirit - that sense of enterprise.
... Kirk, "I, Mudd," stardate 4513.3..