Hi, all. We have a couple of small clusters on dual proc Xeon 32/64 bits and dual core dual proc Xeon on Gigabit and Cent OS 4.4, Intel MKL and FC.
Things are ok but performance is lackluster on the dual core dual proc Xeon machines.
I would like to improve VASP performance and optimize VASP. Reading the manual helps but at times I could not find answers so I would like to ask a few questions here:
How relevant PROC_GROUP in the current state of my settings? The majority of our jobs have been running on 4 or 8 procs. What is the recommended value for PROC_GROUP?
What are the recommended values for MPI_BLOCK for serial and parallel codes?
Our dual core dual proc machines has 8 Gig of RAM and this amount seems to hold up very well. It's appeared to me that cache size of the cores (4 MB) is a big boost to VASP performance. Besides cache size, front size bus speed, CPU, disk, network connection, what else I should pay more attention to improve VASP performance? Is it just helpless to run VASP on dual core dual proc on Gig connection? Should I upgrade the connection to Infiniband or similar?
Thank in advance.
Compilation parameters
Moderators: Global Moderator, Moderator
-
midair77
- Newbie
- Posts: 11
- Joined: Mon Apr 02, 2007 11:32 pm
Compilation parameters
Last edited by midair77 on Tue Apr 03, 2007 12:50 am, edited 1 time in total.
-
admin
- Administrator
- Posts: 2921
- Joined: Tue Aug 03, 2004 8:18 am
- License Nr.: 458
Compilation parameters
the optimum choice of the parameters PROC_GROUP, MPI_BLOCK,... has to be found by benchmark tests made on your machine (with your specific MPI installation)
concerning the communication hardware: Due to as small part of the code which cannot be parallelized (about 3 %, all in all) it is sufficient to use fast Gbit networks.
concerning the communication hardware: Due to as small part of the code which cannot be parallelized (about 3 %, all in all) it is sufficient to use fast Gbit networks.
Last edited by admin on Mon Apr 16, 2007 11:18 am, edited 1 time in total.