Tom Elken wrote:
Which profilers can
benefit from all this info?
We have found Oprofile to be a useful text-oriented tool:
http://oprofile.sourceforge.net/about/
From the Overview on this page:
"OProfile is a system-wide profiler for Linux systems, capable of profiling all
running code at low overhead. OProfile is released under the GNU GPL.
It consists of a kernel driver and a daemon for collecting sample data, and
several post-profiling tools for turning data into information.
OProfile leverages the hardware performance counters of the CPU to enable profiling
of a wide variety of interesting statistics, which can also be used for basic
time-spent profiling. All code is profiled: hardware and software interrupt
handlers, kernel modules, the kernel, shared libraries, and applications."
-Tom
Yes, Oprofile is a fantastic switch to turn on for profiling the entire
system. Now, last time I tried to use it it totally crashed my system. I
had to concentrate on other stable means of profiling at this point.
Note that kTAU is also meant to profile in-kernel execution and seems to
have some fine grained options (not from experience, from reading the
docs ;P). I'm more centered on profiling the applications rather than
the entire system...for the moment, oprofile is still on the list.
Furthermore, oprofile is not "parallel processing" centric, the TAU
suite is and this means that all tools are arranged as to consolidate
and visualize data in that context (check the 3D screen shots).
This said, what I really like is the integration TAU brings, being able
to glue multiple sources of profiling information into a unified format
coupled with GUI visualization tools. The only lacking tool, for the
moment, is a Trace analyzer (although it supports reading/converting
from and to other tracing tool formats such as VTune).
Eric
PS: I am in no way affiliated to the TAU team...btw ;)
This monster suite (free!):
http://www.cs.uoregon.edu/research/tau/docs.php
Which I have been pushing around for a while and it's quite impressive
IMHO. Weird thing is I seem to be the only one interested in
profiling and
metrics and all that gunk around here. Guess that will change
now that it
is actually possible to _measure_ the FLOPS your application _truely_
consumes thanks to these counters (I mean, for us laymans
that never had
access to such cool hardware :P )
toon
Eric
_______________________________________________
Beowulf mailing list, [email protected]
To change your subscription (digest mode or unsubscribe)
visit http://www.beowulf.org/mailman/listinfo/beowulf
_______________________________________________
Beowulf mailing list, [email protected]
To change your subscription (digest mode or unsubscribe) visit
http://www.beowulf.org/mailman/listinfo/beowulf
