Re: [ptp-dev] Debugging on BlueGene

[Greg Watson <g.watson@xxxxxxxxxxxx>]

Wyatt,

Debugging seems to work reasonably well using the interactive PBS RM. We have this working reliably on Forge, though I don't think it's been tested on other systems.

The main problem is getting the debugger to work on BG/P. If you could do this, you'd make a lot of people happy :-).

Currently, the sdm assumes a 1:1 ratio of sdm processes to application processes. Each sdm starts a gdb session that is used to control the application process for debugging. The application must be started under the control of the debugger as there is no way to attach to a running application currently (I'm not sure if this is required for BG/P).

The BG/P architecture is different to a cluster, in that the compute nodes do not run a full operating system. A number of compute nodes are connected to an I/O node which does run a full O/S, so the sdm process would need to be able to control multiple application processes rather than a single one. I believe there's a version of gdb that could be run on the I/O that could act as this multiplexer, but the sdm would need to be modified to be able to handle it. I don't think it would be a huge change, but would require a fairly good technical understanding of BG/P and the sdm.

Let me know if you'd like to discuss in more detail.

Cheers,
Greg

On Feb 7, 2012, at 8:48 PM, Wyatt Spear wrote:

> Greetings,
> 
> Now that we are approaching a functional resource manager for BG/P's PBS system we would like to use Eclipse/PTP to do parallel application debugging on the platform.  What is the status of debugging support through (presumably interactive) LML/PBS resource managers?  Is there anything that needs work that we could be looking at on the Oregon side?
> 
> Regards,
> Wyatt Spear
> _______________________________________________
> ptp-dev mailing list
> ptp-dev@xxxxxxxxxxx
> https://dev.eclipse.org/mailman/listinfo/ptp-dev