[mpich-discuss] MPD Auto

[Dave Goodell]

On Jan 7, 2011, at 12:43 PM CST, Roberto Neves Gonzaga wrote:

> My MPICH2 is working, but I wonder if there is any way to automatic recognition of the machines in the cluster without having to use the command mpiboot. What happens is this: every time you restart the machine have to use the command mpdboot-n 5-f mpd.hosts to recognize the 5 machines in my cluster, not doing so can only run on the master. I am using Fedora 13. 

This sort of question comes up all the time, so I've added a FAQ entry about it: http://wiki.mcs.anl.gov/mpich2/index.php/Frequently_Asked_Questions#Q:_I_don.27t_like_.3CWHATEVER.3E_about_mpd.2C_or_I.27m_having_a_problem_with_mpdboot.2C_can_you_fix_it.3F

Basically, just use hydra instead.  You'll still need to specify the hostlist somehow, either a statically created host file that you reuse (and possibly indicate via HYDRA_HOST_FILE), or by using a resource manager that hydra supports (search for "rmk" in this page): http://wiki.mcs.anl.gov/mpich2/index.php/Using_the_Hydra_Process_Manager

-Dave