[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
Re: [ptp-dev] Trace View of the g-Eclipse project
|
Hi
> Thomas,
> Sorry for the delay in responding; I was at a review meeting all last week.
> How much of this pre-reqs running on your grid infrastructure? Hopefully it
> can also be of use to other parallel machine (more typical PTP user)
> scenarios.
Most of the functionality of the trace view has no strong dependency to the
rest of g-Eclipse. The trace part is quite independent and the debugging part
bases on CDT.
If you have no Grid available at all you still can open trace files and have a
look at the recorded data. The debugging features in the trace view are
usable for MPI programs started using the glogin tool
(http://www.gup.uni-linz.ac.at/glogin/) right now (there is an own launch
configuration type for that in g-Eclipse), this requires that you have glogin
installed on a Grid node which allows to submit jobs using the Globus
middleware (as Globus 2 jobs). But since the debugging functionality is
basing on CDT, I think it should be possible to extend it so that it can be
used programs which have been started using PTP. (It might also be nice to
have some glogin resource manager which allows to plug in glogin based job
submission into PTP, but I guess this is something that needs a little more
thought.)
> How is the trace gathered? Perhaps we could also integrate it usin the PTP
> Performance Tools Framework (PTFw)?
>
> Wyatt and I are in the process of documenting this further; I should have
> an outline at least there later this week. It should be available at
> http://wiki.eclipse.org/PTP/designs/perf_tools_framework
> Both Wyatt and I have been out of town etc. but we know this needs to be
> done so others can use it.
>
> The PTFw would make launching/trace Gathering available from a profile
> launch, and give a callback to perhaps display the trace view.
> Or perhaps your existing infrastructure has some of this?
At the moment the traces come from prerecorded program runs. These traces can
come from NOPE or a tool which generates OTF traces (the OTF support is still
in a very early state). To display data in the traceview there needs to be
some trace provider which returns an instance of the ITrace interface which
is then used by the traceview to get the trace data. This info could also be
returned by something that does the data collection online, this might still
need a bit of work, but is possible basically. Other plugins have the
possibility to pass an ITrace to the traceview so it would be possible to add
a trace for example during program launch.
Regards,
Thomas