- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
So, I setup a coprocessor (5110p card with B1 stepping) to run MPI code based on the very useful info here http://software.intel.com/en-us/articles/mpi-specific-files-for-intel-xeon-phi-what-is-needed. The first step is to copy over the MPI libraries, etc, using:
[user@host] $ scp /opt/intel/impi/<version>/mic/bin/* node0-mic0:/bin
[user@host] $ scp /opt/intel/impi/<version/mic/lib/* node0-mic0:/lib64
Unfortunately, I mistyped and left of the "mic" subdirectory and instead copied:
[user@host] $ scp /opt/intel/impi/<version>/bin/* node0-mic0:/bin
[user@host] $ scp /opt/intel/impi/<version/lib/* node0-mic0:/lib64
It didn't take very long before I realized the mistake and copied over the correct files; however, many other files in both .../bin and .../lib were copied over, possibly overwriting files that were in /bin and /lib64 on the coprocessor. The workstation is running CentOS 6.3 (x64), and the latest version of Intel Cluster Studio xe is installed.
Could anyone post the files that should be in /bin and /lib64 so that I can at least check to see if I overwrote critical files?
BTW, I can run MPI code on the coprocessor from the coprocessor (as well as on the host from the host), but when I try to run MPI code on the coprocessor from the host, there appears to be a communication problem from the coprocessor to the host, and I get the errors:
[proxy:0:0@Axial-mic0.localdomain] HYDU_sock_connect (./utils/sock/sock.c:241): unable to connect from "Axial-mic0.localdomain" to "172.31.1.254" (No route to host)
[proxy:0:0@Axial-mic0.localdomain] main (./pm/pmiserv/pmip.c:353): unable to connect to server 172.31.1.254 at port 38444 (check for firewalls!)
At this point I don't know if this is due to my OS butchering on the coprocessor, but I don't think so.
Thanks All!
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
All the files you copied over will go away with sudo service mpss restart
It's also a good way to clear out any hung processes associated with MPI.
There were a couple of not well advertised changes a few months ago in launching MPI from host side. Read about the new environment variables which were added to support MIC and I believe recommendation to use mpirun rather than previous mpiexec.hydra.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Excellent. I was wondering how the ram drive and flash rom worked (but hadn't given it much thought).
I'll look for the newer MPI info (but if you happen to have the link, that would be great).
Thanks again

- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page