Jeff Squyres wrote:
>On Nov 8, 2005, at 10:53 AM, Prakash Velayutham wrote:
>
>
>
>>I am currently working on getting some support for dynamic scheduling
>>in
>>Torque. This means not just asking for a fixed number of resources
>>during submission of a job, but later requesting to add more resources
>>to the job or giving up some when not needed. I did bring it up on this
>>list several months back. Currently lamgrow errors out when called from
>>inside of a PBS setting due to support not being available from
>>PBS/Torque. How about looking at this feature of lamgrow from TM
>>interface in LAM as it might be possible to include support in Torque?
>>
>>
>
>Are you saying that the Torque guys are going to extend the TM
>interface to support this kind of functionality?
>
>
They were (Dave Jackson and Daniel Reese) interested when I proposed it
initially. But I don't get much help or direction from them as they seem
to be really busy. Actually I am doing this as a part of my Master's
thesis in Univ. of Cincinnati. Currently I am doing these on my own
against Torque-1.2.0p5 sources. Maybe Torque guys will be willing to
take a look when it is in more or less working condition and if I just
give them a patch set against the current version. That is probably a
little into the future.
>
>
>>Could someone please let me know what is the best way to get this
>>support in LAM's sources?
>>
>>
>
>To be honest, we'd probably be much more interested in integrating this
>into Open MPI rather than LAM. LAM is essentially in a maintenance
>mode; we're not really adding new features.
>
>In some ways, it would be a bit easier to add this to Open MPI because
>-- at least as of yet -- Open MPI does not have persistent daemons like
>LAM does. However, Open MPI's runtime is much more flexible and
>adaptable than LAM's, and is therefore a bit more complicated. So it
>would probably take some work.
>
>
I would very much appreciate any help from your side to get this support
into Open MPI (of course, when Torque is ready). Do you think it is
possible for me to get some help to integrate this and test it at my site?
>But if the Torque guys are going to extend the TM interface, we would
>very much like to be involved in those conversations. We have a lot of
>gripes about the TM interface and would like to have them fixed. :-)
>
I don't know about their interest in fixing all the issues with the TM
interface. I just want to get the TM interface support for MPI-2 dynamic
process management feature. Maybe in the process I might learn enough of
Torque to fix those other things you are interested and get those done
into Torque source.
Please let me know.
Thanks,
Prakash
|
