[mpich-discuss] Hydra round robin questions

Pavan Balaji balaji at mcs.anl.gov
Wed Feb 16 09:36:28 CST 2011


Hi Saurabh,

On 02/16/2011 08:46 AM, Saurabh T wrote:
> I made a hosts file like this:
> name0:2 binding=rr
> name1:4 binding=rr
> name3:1 binding=rr
>
> to utilize various machines with various number of cpus at our site.
> However when I run mpich, the ranks are assigned:
> [name0,name0,name1,name1,name1,name1,name3, {repeat}]. I want them to
> be assigned: [name0,name1,name3,name0,name1,name1,name1, {repeat}].
> How do I do this?

In your host file, you specifically asked for 2 processes to be launched 
on node0, then 4 on node1, etc. So Hydra is doing exactly that. You can 
control how many processes you want using the :x option after the host 
name. The following might be what you want?

name0
name1
name3
name0
name1:3

> Second, how do I always make rank 0 start on the machine where mpirun
> was invoked?

Hydra just follows nodes in the order you listed. You can put the 
launching node first in the host list.

  -- Pavan

-- 
Pavan Balaji
http://www.mcs.anl.gov/~balaji


More information about the mpich-discuss mailing list