From: Tristan Croll (tristan.croll_at_qut.edu.au)
Date: Fri Nov 08 2013 - 22:11:41 CST
Hi all,
I'm running a moderately-sized (~380k atom) simulation in NAMD 2.9 on the new national compute server here in Australia (http://nf.nci.org.au/facilities/fujitsu.php). I'm seeing some strange behaviour in terms of scaling, and was hoping to get some ideas on where it might be coming from. The raw numbers are below. The most obvious problem is the enormous slowdown past 384 cores; my contact at the national facility tells me this is likely an OpenMPI problem which they've also seen in other packages once message sizes get too small. This isn't really my concern, since I typically stick to 256 cores. What worries me more is the ~50% loss in efficiency between 16 and 128 cores, when most benchmarks I've seen on other systems show near-linear scaling in this range. Is there anything I can do from my end to improve on this?
Thanks,
Tristan
Cores CPU-seconds/step s/step
16 3.74 0.234
32 4.16 0.13
64 4.99 0.078
128 5.63 0.044
256 5.89 0.023
384 5.76 0.015
512 16.90 0.033
768 82.94 0.108
1024 135.17 0.132
Tristan Croll
Lecturer
Faculty of Health
Institute of Health and Biomedical Engineering
Queensland University of Technology
60 Musk Ave
Kelvin Grove QLD 4059 Australia
+61 7 3138 6443
This email and its attachments (if any) contain confidential information intended for use by the addressee and may be privileged. We do not waive any confidentiality, privilege or copyright associated with the email or the attachments. If you are not the intended addressee, you must not use, transmit, disclose or copy the email or any attachments. If you receive this email by mistake, please notify the sender immediately and delete the original email.
This archive was generated by hypermail 2.1.6 : Wed Dec 31 2014 - 23:21:53 CST