From: Dow_Hurst (dhurst_at_mindspring.com)
Date: Sat Sep 22 2007 - 18:28:17 CDT
Carsten,
would you be more specific about your statement? I would think that you would see a big difference in performance between Infiniband and Ethernet. How many nodes, cores, and IB cards are you talking about? Certainly scaling should be much different beyond 16 cores as Gig-E usually starts dying about that point.
Note on the performance benchmark page how different the Gig-E versus Infiniband scaling is.
Have you compiled charm++ and NAMD with the IB MPI?
Thanks,
Dow
-----Original Message-----
>From: Carsten Olbrich <ocarsten_at_googlemail.com>
>Sent: Sep 22, 2007 7:44 AM
>To: Namd Mailing List <namd-l_at_ks.uiuc.edu>
>Subject: namd-l: NAMD on dual quad-core Xeon's with Infiniband backbone
>
>Hi Stephen,
>
>we have a quad core Xeon system with Infiniband and Ethernet.
>We observed no performance differences between network types.
>But you should use the Linux-amd64 build. This is about twice as fast
>as the Linux-i686 build. Also the option +giga increased the
>performance.
>
>Regards,
>
>Carsten
>
>On 9/21/07, Stephen M. Dutz <smdutz_at_csupomona.edu> wrote:
>>
>>
>>
>>
>> Hi all,
>>
>>
>>
>> I've searched the message board and the wiki for posts on running NAMD on
>> dual-quad core Xeon systems with an Infiniband backbone, however haven't
>> found much data. I'm currently running the Linux-i686 build and in the
>> process of compiling the source code to see if this speeds things up at all.
>> I was wondering if anyone out there has tried running NAMD on a similar
>> system, and if so what was your methodology for maximizing performance, and
>> what type of scaling do you observe.
>>
>>
>>
>> Thanks in advance,
>>
>>
>>
>> Steve
>>
>>
>
No sig.
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:45:17 CST