[Beowulf] RE:CCL:G: parallel G03 in Linux cluster

Don Kinghorn kinghorn at pqs-chem.com
Mon Mar 20 09:04:37 PST 2006


Hi Debasis,

I think I can answer your questions;

1)  G03 runs well on dual core Opterons/Athlon64  I would recommend Opterons 
and other server grade components including ECC memory unless you are on a 
tight budget. 

If your budget is tight and you are going to be doing mostly DFT type calcs 
you should get good value and performance with dual-core. If you are doing 
correlated methods then you may wany to consider single core processors since 
you will need access to fast scratch disk space (two cores will compete for 
the disks slowing down your jobs)

2) Yes, you will need to get a Linda license to go parallel over a network.

3)  You would be OK with Intel but I can tell you that the people at Gaussian 
prefer AMD and SuSE linux. Also, I believe you will likely get your best 
price performance with AMD.

4) They grudgingly support RedHat but they use SuSE in house You should be OK 
with SuSE 10.0  ( SuSE 10.1 will out soon but go with 10.0 for g03 )

5)  If you have trouble with the install drop we a line. I can send you a 
simple bash environment script to drop in /etc/profile.d so you don't have to 
mess around with csh stuff.  I recommend that you get a binary build rather 
than trying compile it yourself.

Best wishes
-Don

> Sent to CCL by: Debasis Sengupta [dxs]=[cfdrc.com]
> Hello Everybody,
> 
> I am planning to buy sixteen AMD64 Athlon dual core machines (with 
> Linux) for running G03 in parallel. They will be connected by ethernet. 
> I never have any experience in running gaussian with Linux cluster.  So 
> my questions are the following:
> 
> 1.  Can G03 be run parallel in the above configuration?
> 2. Do you need to have Linda for running parallel in the above 
> configuration?
> 3. Do you recommend any other processor?
> 4. Which brand of Linux is good for running G03 parallel?  
> 5. Pals sharer your experience with  running G03 parallel in Linux 
> cluster (like tricks, problems, failure etc).
> 
> Thanks in advance
> Debasis
-- 
Dr. Donald B. Kinghorn Parallel Quantum Solutions LLC
http://www.pqs-chem.com




More information about the Beowulf mailing list