[mpich-discuss] MPI on a cluster of dual-CPU machines

[Rajeev Thakur]

The default channel in 1.0.8 is ch3:sock, which will use TCP for
communication even within a node. If you configure with
--with-device=ch3:nemesis, it will use shared memory for communication
within a node and TCP across. In 1.1, nemesis is the default. The way to
launch processes and MPD configuration hasn't changed between 1.0.8 and
1.1.

Rajeev


> -----Original Message-----
> From: mpich-discuss-bounces at mcs.anl.gov 
> [mailto:mpich-discuss-bounces at mcs.anl.gov] On Behalf Of Pavan Balaji
> Sent: Friday, July 17, 2009 7:09 PM
> To: mpich-discuss at mcs.anl.gov
> Subject: Re: [mpich-discuss] MPI on a cluster of dual-CPU machines
> 
> 
> > Just one more question though. You say that the latest 
> release detects 
> > this configuration. I am using mpich2-1.0.8. Does this version too 
> > detect the presence of dual CPUs automatically?
> 
> Not with the default configuration. You'll need to configure 
> with --with-device=ch3:nemesis for that. But, it's still 
> advisable that you move to 1.1, since ch3:nemesis has had 
> quite a few fixes and major improvements between 1.0.8 and 1.1.
> 
>   -- Pavan
> 
> --
> Pavan Balaji
> http://www.mcs.anl.gov/~balaji
>