Opened 9 years ago

Closed 9 years ago

#555 closed defect (worksforme)

OpenMPI appears to not work in Liberation

Reported by: fitz Owned by: skylar
Priority: major Milestone:
Component: Both Version:
Keywords: Cc:
Blocked By: Blocking:
Estimated Hours: 0 Total Hours: 0

Description

But only with -np >= 3. Things hang in an apparent deadlock.

Change History (4)

comment:1 Changed 9 years ago by fitz

On the headnode, OpenMPI is selecting the wrong interface to communicate over.

The quick fix is to find the interface that has the 192.168.3.1 IP and disable it, e.g.:

$ ip addr | grep 192.168.3.1

inet 192.168.3.1/24 brd 192.168.3.255 scope global eth2:1

$ sudo ifdown eth2:1

This is, of course, assuming eth2 is already set up with a different IP as per 1.

A more permanent fix might be to disable that alias if the user specifies an alternate IP during boot or reset-network.

comment:2 Changed 9 years ago by skylar

got VMs liberated and snapshotted

PXE nodes ready

gotta go to convention center now

comment:3 Changed 9 years ago by skylar

  • Owner set to skylar
  • Status changed from new to assigned

comment:4 Changed 9 years ago by skylar

  • Resolution set to worksforme
  • Status changed from assigned to closed

I think was a false alarm

Note: See TracTickets for help on using tickets.