RE: [bgl-discuss] MPI node 0 on bg/l

["Lee, Hyong" <hlee@xxxxxxxxxxxxxxxxxxxxx>]

Title: RE: [bgl-discuss] MPI node 0 on bg/l

Hi Ray & Rob,

Thanks for your comments.  I guess I should have been more explicit, the integrator is the numerical solver.  It is mostly a black box at the script level, so hard to debug without poking around in the source, which I am starting to do.  At least I know to suspect something pgenesis is doing rather than some heterogeneity in the bg/l nodes.

I've looked at some of the bg/l docs, though by no means all; I have noticed some of these limitations, so I will look to see if, for example, pgenesis tries to spawn additional processes on node 0.  I've had to hack a bit of the tty and diskio interface to get it to compile, so there may be some problems there as well.

Are there any special considerations when debugging on bg/l?  What are the main tools for this?

Hyong Lee
Software Engineer
UofC Hospitals



-----Original Message-----
From: Rob Ross [mailto:rross@xxxxxxxxxxx]
Sent: Tue 9/11/2007 4:59 PM
To: Lee, Hyong
Cc: discuss@xxxxxxxxxxxxxxx
Subject: Re: [bgl-discuss] MPI node 0 on bg/l

Hi Hyong,

Maybe you should describe what an "integrator" does?

All the nodes of BG/L are "special" as compared to Linux nodes; lots of
common UNIX functionality isn't there. Some system calls will link
correctly but return ENOSYS. But I'm not sure that this is relevant.

Regards,

Rob

Lee, Hyong wrote:
> Hi all,
>
> I'm trying to get pgenesis, a neural simulation environment, to work on
> BG/L.  It consists of an interpreter for a csh-like scripting language
> and an integrator, among other things, all in the same binary.
>
> It compiles (with some minor modifications) and runs, and the script
> interpreter runs fine on all nodes; however, when we try to call the
> integrator, it seems to hang waiting for messages from node 0.  The same
> scripts work on Jazz.
>
> Is there anything special about MPI node 0 on BG/L?  If so, is there
> some documentation I can look at?  In the mean time, I am trying to see
> if our model will run if we avoid integrating on node 0; of course, we
> would like to understand the problem so we have some confidence in any
> results we get with the workaround.
>
> Thanks for your help!
>
> Hyong Lee
> Software Engineer
> UofC Hospitals
>

This email is intended only for the use of the individual or entity to which it is addressed and may contain information that is privileged and confidential. If the reader of this email message is not the intended recipient, you are hereby notified that any dissemination, distribution, or copying of this communication is prohibited. If you have received this email in error, please notify the sender and destroy/delete all copies of the transmittal. Thank you.