How to run the ROMS to MPI in XeonPhi

Discussion of how to use ROMS on different regional and basin scale applications.

Moderators: arango, robertson

Post Reply
Message
Author
kissrang
Posts: 5
Joined: Thu Jun 11, 2015 5:13 pm
Location: YoungDong University

How to run the ROMS to MPI in XeonPhi

#1 Unread post by kissrang »

Hello

I've got eight XeonPhi system.
I was running a successful ROMS to MIC mode.
I want to run at the same time as the ROMS eight XeonPhi.
I used the intel mpi.
I've created a machinefile mics.
<Mics>

mic0: 24
mic1: 24
# Mic2
# Mic3
# Mic4
# Mic5
# Mic6
# Mic7

Only two XeonPhi've entered for testing.
And
: ~> mpiexe.hydra -machinefile mics -n 48 ./oceanM-mic ROMS / External / ocean_upwelling.in

phi-test-mic0: SCM: 51fe: e5061b40: 1119 us (1119 us): open_hca: ibv_get_device_list () failed
phi-test-mic0: SCM: 51fe: e5061b40: 1155 us (1155 us): open_hca: ibv_get_device_list () failed
phi-test-mic0: CMA: 51fe: e5061b40: 1148 us (1148 us): open_hca: getaddr_netdev ERROR: No such device. Is ib0 configured?
phi-test-mic0: CMA: 51fe: e5061b40: 1488 us (1488 us): open_hca: getaddr_netdev ERROR: No such device. Is ib1 configured?
phi-test-mic0: SCM: 51fe: e5061b40: 1097 us (1097 us): open_hca: ibv_get_device_list () failed
phi-test-mic0: SCM: 51fe: e5061b40: 1118 us (1118 us): open_hca: ibv_get_device_list () failed
phi-test-mic0: SCM: 51fe: e5061b40: 979 us (979 us): open_hca: ibv_get_device_list () failed
phi-test-mic1: SCM: 20ea: 372a5b40: 1118 us (1118 us): open_hca: ibv_get_device_list () failed
phi-test-mic1: SCM: 20ea: 372a5b40: 252 us (252 us): open_hca: ibv_get_device_list () failed
phi-test-mic0: SCM: 51fe: e5061b40: 900 us (900 us): open_hca: ibv_get_device_list () failed
.
.
(Many error messages.)
.
.
many
What's wrong?


Thank you.

gnayar

Re: How to run the ROMS to MPI in XeonPhi

#2 Unread post by gnayar »

Hi kissrang,
The best place to post Xeon Phi related issues is to post in https://software.intel.com/ forums. I found a post with a similar issue with WRF which has some good points for you to try.

Here is the post: https://software.intel.com/en-us/forums/topic/542161

You must make sure that you have the /opt folder in the host mounted on both the mics, source the host mpivars.sh since you are launching the app from the host, make sure you have the driver installed for the fabric you are trying to use, set I_MPI_MIC to 1...

You can try setting I_MPI_FABRICS to shm:tcp to check if it is driver related.

-gopal

Post Reply