Hello,
With darshan 2.2.0, darshan-job-summary.pl hangs when used on data
generated from an application doing I/O operations on a filesystem that
has the underscore character in its name (/home_nfs for example).
Actually, pdflatex is the one hanging because of that underscore
character.
To solve this issue, just add "usepackage{underscore}" into
darshan-job-summary/share/summary.tex.
=============================================
--- darshan-job-summary/share/summary.tex 2012-06-26
15:03:46.000000000 +0200
+++ darshan-job-summary/share/summary.tex.patched 2012-06-26
15:03:33.000000000 +0200
@@ -6,6 +6,7 @@
\usepackage{lastpage}
\usepackage{subfigure}
\usepackage{multirow}
+\usepackage{underscore}
%
% GET THE MARGINS RIGHT, THE UGLY WAY
%
=============================================
Thanks,
Mouhamed GUEYE
HPC Engineer
Phone +33 (0)4 76 29 74 91
Bull, Architect of an Open World
http://www.bull.com
Bull World Web TV -> http://www.bull-world.com/
Hello,
I am trying to profile OpenFOAM (http://www.openfoam.com/) with Darshan.
This seems to work pretty good, although I have encountered a little
problem that I hope you could help me with:
I am trying to write the data from each process to /dev/shm to reduce
the total filesystem IO (witch seems to be a bottleneck in some cases).
The problem is that Darshan does not capture this data, and report 0 MB
written to an unknown filesystem. See attached PDFs. The case is
identical except that in the first case (Normal.pdf) all processes
writes data to /home, and in the last case (DEV-SHM.pdf) only a few
files are read from /home, and the rest is read and written from /dev/shm.
Does any of you have some ideas why Darshan does not capture the data
written to/from /dev/shm?
I have tried doing this on my workstation, with another version of
OpenFOAM (newer), and the same thing happens. OpenFOAM and Darshan is in
both cases compiled with the same compiler, Intel and GNU respectivly.
Best regards,
H�kon Strandenes
Norwegian University of Science and Technology
Hi all,
There is a pre-release version of Darshan 2.2.1 available at
http://www.mcs.anl.gov/darshan/download/. As with any pre-release, this
2.2.1-pre1 version should not be used in production. We may make
changes before the final 2.2.1 release based on testing. We would love
to hear feedback from anyone who would like to try it out, though.
This release does a better job of tracking concurrent access from
multi-threaded applications, which should be helpful for hybrid
applications that use a combination of MPI and threading (or OpenMP).
The other big change is significantly improved Cray support, although we
intend to polish this further in upcoming releases.
thanks,
- the Darshan team
darshan-2.2.1-pre1 Changelog
=============
* significant improvements to how counters are handled in multi-threaded
applications
* initial (rough) documentation for using Darshan in the default Cray
programming environment with static linking
* bug fixes:
- workarounds for various runtime problems with cuserid() and stat() in
Cray environments
- build problems with darshan-utils on some versions of OSX
- accurate shared file statistics for libraries that use deferred opens
