[GE users] Help: PE Question

Chris Dagdigian dag at sonsorol.org
Sat Sep 22 14:30:50 BST 2007


Good to see more bioinformatics types on this list!

The following blog links may be useful in filling in background info:

http://gridengine.info/articles/2005/09/19/parallel-environments-pes- 
loose-vs-tight-integration

http://blogs.sun.com/templedf/entry/pe_tight_integration


The long and the short of it is that when you have a MPI environment  
that you want to play nicely with SGE then the SGE scheduler needs to  
be involved in one or more of the following tasks:

1. Picking the specific hosts that the parallel job is dispatched onto
2. Starting/stopping/controlling the parallel tasks themselves


Task #1 is essential and has to happen no matter what -- if you have  
a 32 node cluster and someone starts a 4-way parallel job, the  
scheduler has to step in and look at the state of the system before  
deciding which 4 CPUs will get the job. This can generally be called  
"loose integration" if you stop at this point.

Task #2 is optional but desirable; the main downside is complexity in  
setup and dealing with the fact that sometimes tight integrations can  
be somewhat application or PE specific. The main benefits is that SGE  
gains visibility and some level of control over the parallel tasks so  
you get better job management and resource accounting.

If you are new to MPI and PE's and have a MPICH program you want to  
get working under SGE then one fairly straightforward path would be:

1. prove that your MPI program and MPI installation environment works  
by itself before you add the complexity of SGE integration (you'd be  
surprised how many people don't bother to do this and end up blaming  
SGE for problems that end up being traced to the MPI install or the  
app itself)

2. setup your PE; go for loose integration

3. Test your MPI app within SGE

4. Decide if you want to go for the tight integration features


-Chris






On Sep 22, 2007, at 8:34 AM, Amy Lee wrote:

> Hi,
>
> I'm a Linux system administrator in the Bioinformatics Department  
> of my university. And I have used SGE to make EMBOSS and PHRAP run  
> more coefficient. However, I still have some questions about PE in  
> SGE.
>
> I've download the SGE PDF documents from the official website. I  
> wonder that whether I can understand the PE part clearly. In fact,  
> I wanna know how to have a configuration with the PE.
>
> I use MPICH 1.4, and there are some parameters /opt/sge/mpi/ 
> startmpi.sh and stopmpi.sh scripts, I wanna know the meanings of  
> them. How to combine the scripts with MPICH?
>
> Thank you very much~
>
> Regards,
>
> Amy Lee
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
> For additional commands, e-mail: users-help at gridengine.sunsource.net

---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list