I think the information was scattered across a few posts, but the union of 
which is correct:

- it depends on the benchmark

- yes, L1/L2/L3 cache sizes can have a huge effect.  I.e., once the buffer size 
gets bigger than the cache size, it takes more time to get the message from 
main RAM
  --> check the output from hwloc's "lstopo" tool to find your cache sizes

- the specific flavor of shared memory used also has a huge effect.  The 
default is copy-in/copy-out, but other shared memory mechanisms are also 
available (e.g., Linux CMA, Linux KNEM, XPMEM)

Does that help?


> On Mar 10, 2016, at 12:25 PM, BRADLEY, PETER C PW 
> <peter.c.brad...@pw.utc.com> wrote:
> 
> This is an academic exercise, obviously.  The curve shown comes from one pair 
> of ranks running on the same node alternating between MPI_Send and MPI_Recv.  
> The most likely suspect is a cache effect, but rather than assuming, I was 
> curious if there might be any other aspects of the implementation at work.
> 
> Pete
> 
>  
> 
> Pete, 
> 
> how did you measure the bandwidth ? 
> iirc, IMB benchmark does not reuse send and recv buffers, so the results 
> could be different. 
> also, you might want to use a logarithmic scale for the message size, so 
> information for small messages is easier to read. 
> 
> Cheers, 
> 
> Gilles 
> 
> On Thursday, March 10, 2016, BRADLEY, PETER C PW 
> <peter.c.bradley_at_[hidden]> 
> wrote: 
> 
> > I’m curious what causes the hump in the pingpong bandwidth curve when 
> > running on shared memory. Here’s an example running on a fairly 
> > antiquated 
> > single-socket 4 core laptop with linux (2.6.32 kernel). Is this a cache 
> > effect? Something in OpenMPI itself, or a combination? 
> > 
> > 
> > 
> > 
> > 
> > [image: Macintosh HD:Users:up:Pictures:bandwidth_onepair_onenode.png] 
> > 
> > 
> > 
> > Pete 
> > 
> > 
> > 
> 
>  
> _______________________________________________
> users mailing list
> us...@open-mpi.org
> Subscription: http://www.open-mpi.org/mailman/listinfo.cgi/users
> Link to this post: 
> http://www.open-mpi.org/community/lists/users/2016/03/28674.php


-- 
Jeff Squyres
jsquy...@cisco.com
For corporate legal information go to: 
http://www.cisco.com/web/about/doing_business/legal/cri/

Reply via email to