mpd failed to boot

[Chao_M_]

Hi all,

When I try to run mpdboot on single node (RHEL7), but I got error message:

• [root@hpc-test intel64]# mpd &
• [1] 3527
• [root@hpc-test intel64]# mpd_uncaught_except_tb handling:
•   <type 'exceptions.IndexError'>: list index out of range
•     /opt/intel/impi/5.0.2.044/intel64/bin/mpd  264  pin_Uni_num
•         if list.index(list) == i:
•     /opt/intel/impi/5.0.2.044/intel64/bin/mpd  1449  pin_Cpuinfo
•         info['cache1'] = pin_Uni_num(info['cache1_id'], info['lcpu'])
•     /opt/intel/impi/5.0.2.044/intel64/bin/mpd  1658  run
•         self.CpuInfo = pin_Cpuinfo(self.PinCase,self.Arch)
•     /opt/intel/impi/5.0.2.044/intel64/bin/mpd  3676  <module>
•         mpd.run()

I use I_MPI_CPUINFO=proc to work around, but I don't know what cause it, and how to fix it correctly?

Thank you for your help.

Thanks,

Chao

[Gergana_S_Intel]

Hi Chao,

Thanks for getting in touch.

The mpd scripts (mpd, mpdboot, mpdallexit, etc.) have been deprecated starting with Intel MPI 5.0 and are no longer shipped with the library.  I would recommend switching over to using the Hydra process manager which has been our default for several releases now.

To use Hydra, just replace your existing scripts with mpirun.  For example, if you had previously done this:

$ mpdboot
$ mpiexec -n 4 ./a.out
$ mpdallexit

You can now simply say:

$ mpirun -n 4 ./a.out

Let me know how it goes or if you have questions.

Regards,
~Gergana

[Chao_M_]

Thank you, Gergana.

So I can use mpiexec.hydra or simply use mpirun instead mpdboot etc.

Thanks,

Chao

[Gergana_S_Intel]

Hi Chao,

Yes, indeed.  The mpirun script uses mpiexec.hydra by default so either is fine.

Let us know if you hit any problem.

All the best,
~Gergana