[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[bgl-discuss] foo

To: discuss@xxxxxxxxxxxxxxx
Subject: [bgl-discuss] foo
From: Ray Loy <rloy@xxxxxxxxxxx>
Date: Wed, 14 Dec 2005 19:02:39 -0600
Cc: jacob@xxxxxxxxxxx
Comments: In-reply-to Pete Beckman <beckman@mcs.anl.gov> message dated "Thu, 27 Oct 2005 07:50:39 -0500."
Delivered-to: discuss-outgoing@bgl.mcs.anl.gov
Delivered-to: discuss@bgl.mcs.anl.gov
In-reply-to: <A5689606-148F-4E3A-8750-BE6C32424E1C@mcs.anl.gov>
References: <A5689606-148F-4E3A-8750-BE6C32424E1C@mcs.anl.gov>
Sender: owner-discuss@xxxxxxxxxxxxxxx

The recent driver update has fixed the CO-mode scaling problem with
POP 1.4.3, as we hoped it would.  Which would suggest the Allreduce
bug was the culprit.


Ray



> From: Pete Beckman <beckman@xxxxxxxxxxx>
> Subject: [bgl-discuss] allreduce performance bug in coprocessor mode
> Date: Thu, 27 Oct 2005 07:50:39 -0500
> 
> While running some benchmarks, we found out that MPI_Allreduce has a  
> serious performance bug in coprocessor mode on our machine (driver  
> 202).  Allreduce for virtual node mode is 100x faster (for twice the  
> number of processors).  Allreduce for co-processor mode is slow.
> 
> The bug is fixed in the next IBM driver release, but in the mean  
> time, realize allreduce will be very slow in coprocessor mode.

Attachment: bgl.png
Description: Binary data

Follow-Ups:
- Re: [bgl-discuss] foo
  - From: Kamil Iskra

Next by Date: Re: [bgl-discuss] foo
Next by thread: Re: [bgl-discuss] foo
Index(es):
- Date
- Thread