[mpich-discuss] credentials for mpiexec -f machinefile

Kenneth Raffenetti raffenet at mcs.anl.gov
Wed Apr 23 16:32:17 CDT 2014


You can control the port range that mpich will use. See this item in the FAQ

 
http://wiki.mpich.org/mpich/index.php/Frequently_Asked_Questions#Q:_How_do_I_control_which_ports_MPICH_uses.3F

Ken

On 04/23/2014 04:26 PM, Jan Balewski wrote:
>
> On Apr 23, 2014, at 5:10 PM, "Antonio J. Peña" <apenya at mcs.anl.gov> wrote:
>
>> No, what I meant was doing "ssh localhost", but now I look at your logs again, it may be a problem of your DNS configuration.
>> mpi-966f395e-bbb1-4a20-8cbc-c10081e91244 seems to be unable to resolve the IP address of mpi-be6bebee-55e3-4901-a5bb-637395ba46f6. Take a look at your /etc/hosts files in both hosts.
>>
>
> Right, I just figured that out as well. When VM's are launched the get assigned this funny names, not matching the DNS names.
> So I have manually adjusted the content of this 3 files  (I run on SL6.5)
>   vi /etc/hostname
> vi /etc/sysconfig/network
> vi /etc/hosts
>
> so the DNS name , public IP, and hostname --fqd  return the same info, e.g.
>
> [cosy11 at oswrk210 mpich-3.1]$ hostname --fqd
> oswrk210.lns.mit.edu
>
> Now the command below  does not crashes  but hangs for many minutes, eventually time-outing.
> So looks like I need to open port  60568 , right?
> After I shut down fire wall on IP=210 it complain abut port   45895
> Next I shut down firewall in IP=207
> and it complained abut new port  35167
>
> Finally I shut down firewall between VM (on the openStack controller) and … IT WORKED:
>
> [cosy11 at oswrk210 mpich-3.1]$ mpiexec -f machinefile2 -n 4 ./examples/cpi
> Process 0 of 4 is on oswrk210.lns.mit.edu
> Process 3 of 4 is on oswrk207.lns.mit.edu
> Process 1 of 4 is on oswrk210.lns.mit.edu
> Process 2 of 4 is on oswrk207.lns.mit.edu
> pi is approximately 3.1415926535899028, Error is 0.0000000000001097
> wall clock time = 0.003602
>
> QUESTION:
> Can you tell me what ports I need to have open on the master & worker VMs for  mpiexec  to distribute work on them?
>
> Jan
>
>
>
>
> _______________________________________________
> discuss mailing list     discuss at mpich.org
> To manage subscription options or unsubscribe:
> https://lists.mpich.org/mailman/listinfo/discuss
>



More information about the discuss mailing list