Hi hui 

thanks for your response. I tried the manual mode on a simply hello world mpi application, it works as expected. 
You are pretty mind-reading. I have two follow up questions, one is that as you said how can we detect launch failures? 
another one is that except the launching part, all other things like fault tolerance, recovery, etc will be all the same under manual mode as other modes, nothing special under manual mode right? 

Thanks
Shuwei

On Fri, Oct 26, 2018 at 11:29 Zhou, Hui <zhouh@anl.gov> wrote:
In fact there is a manual mode in hydra. Try this:

    mpiexec -n 2 -launcher=manual -host=A,B date

It will print out the launch command lines and wait (very patiently). Then if you paste the proxy command line on your local host or any host on your network, it will continue. 

I guess that is all the pipelines that you can use to create your own wrapper scripts to work for custom environment. 

— 
Hui

On Oct 26, 2018, at 8:42 AM, Zhou, Hui <zhouh@anl.gov> wrote:

Having a manual mode of launching proxies sounds very interesting. I don’t think currently hydra supports it though. 

The manually launched proxies need establish communication back to mpiexec.hydra — I suspect the mechanism already exist. I guess the next question is how to reliably detect launch failures other than having mpiexec.hydra hanging forever — maybe that is a valid option. 

Hui Zhou

On Oct 23, 2018, at 11:44 PM, Shuwei Zhao <shuweizhao1991@gmail.com> wrote:

Hi, 

I'm trying integrate MPICH with network submission tool called RTDA(RunTime Design Automation - www.rtda.com), I use the mpich-3.2.1a but looks like that mpich doesn't detect the resource manager and launcher, mpich cannot distribute jobs as expected.

Does mpich support interact with RTDA resource manager and launcher? 

Since the hydra process manager did tight integration with SGE, LSF, SLURM, PBS, etc. I was thinking for platform it doesn't support - if there is way to submit jobs with loose integration? 
( loose integration means that we do the job submission self and run hydra_pmi_proxy on each allocated node manually, instead of mpich finished everything under the hood.
)

Thank you very much,
Shuwei
_______________________________________________
discuss mailing list     discuss@mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss

_______________________________________________
discuss mailing list     discuss@mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss

_______________________________________________
discuss mailing list     discuss@mpich.org
To manage subscription options or unsubscribe:
https://lists.mpich.org/mailman/listinfo/discuss