[SimGrid-user] IPM with SIMGRID
RAMAMONJISOA, CHARLES EMILE
C.RAMAMONJISOA at AFDB.ORG
Thu Nov 7 09:22:49 CET 2013
I am coming back on this issue. As per your advice, I've used the -cfg and -trace-resource directives with smpirun and I have got the trace file. And then, I have also finally installed Pajé and VITE on my virtual machine but unfortunately, despite I have increased to 12 Gb the memory of my machine, I cannot display and visualize the trace file content. With Pajé, the screen (and all menus) are frozen and with VITE, I can access very slowly some of the menu options but the display screen remains unreadable.
Any clue on this ?
De : simgrid-user-bounces at lists.gforge.inria.fr [mailto:simgrid-user-bounces at lists.gforge.inria.fr] De la part de Frédéric suter
Envoyé : mercredi 9 octobre 2013 09:23
À : simgrid-user at lists.gforge.inria.fr
Objet : Re: [SimGrid-user] IPM with SIMGRID
In addition to what Augustin said, you can have a look at
it may help you to find what you need.
Le 09/10/2013 10:18, Lucas M. Schnorr a écrit :
On 10/09/2013 04:24 AM, RAMAMONJISOA, CHARLES EMILE wrote:
I am looking for a tracing software when running my program in simgrid with SMPI. My concern is particularly to try to evaluate the communication time spent by processes between each node in a complex and heavy platform configuration. I have seen IPM (Integrated Performance Monitoring) but it seems done only for MPI not for SMPI.
What you mean by communication time? If it is only the time spent on blocking operations such as MPI_Send, MPI_Recv, etc, the tracing that is already implemented in SMPI is enough (activated through --cfg=tracing/smpi:1). If you need the time spent on network links that SMPI simulates through SimGrid's simulation kernel, you'll need only the uncategorized resource tracing (--cfg=tracing/uncategorized:1), which registers the consumption for every resource (host and link). So, with the former you'll get an application-level tracing based on the application communication pattern and topology. While with the latter, you'll get a resource-level tracing based on the network topology. You can, of course, trace both at the same time and let your analysis do the correlation.
What you mean by a complex and heavy platform configuration? If you scale up to some thousands, I believe the existing tracing infrastructure is enough. Let us know if not.
A while ago we implemented the PMPI interface in SMPI, so we can let SMPI cooperate with some external tracing software such as IPM. I'm not sure if that is currently possible.
Simgrid-user mailing list
Simgrid-user at lists.gforge.inria.fr<mailto:Simgrid-user at lists.gforge.inria.fr>
L'imagination est une ivresse, la réalité est sa gueule de bois.
Imagition is an inebriation, reality is its hangover.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Simgrid-user