Some Eclipse Foundation services are deprecated, or will be soon. Please ensure you've read this important communication.

Bug 349206

Summary:

Startup errors for PE application using PE resource manager

Product:

[Tools] PTP

Reporter:

David Wootton <drwootton>

Component:

RM.PE

Assignee:

Project Inbox <ptp-inbox>

Status:

RESOLVED FIXED

QA Contact:

Severity:

normal

Priority:

CC:

beth

Version:

4.0.7

Target Milestone:

---

Hardware:

OS:

Unix All

Whiteboard:

Attachments:

Description	Flags
Patch to fix problem running PE application	none
Patch to fix problem running PE application	none

Description David Wootton

2011-06-13 11:58:28 EDT

If I attempt to run a PE application using the PE resource manager on an AIX or Linux system using the internal high speed network on the remote cluster, I get messages stating LAPI initialization failed at the point where the application calls MPI_Init. This is due to the PE proxy forcing MP_RESD=no when LoadLeveler is not being used.

Comment 1 David Wootton

2011-06-13 13:27:42 EDT

Created attachment 197905 [details]
Patch to fix problem running PE application

Comment 2 David Wootton

2011-06-13 13:28:03 EDT

Created attachment 197906 [details]
Patch to fix problem running PE application

Comment 3 David Wootton

2011-06-13 13:30:00 EDT

Modified proxy so that the MP_RESD environment is only set to MP_RESD=yes when LoadLeveler is used. In the non LoadLeveler case, the usr should be able to specify the setting.

Also fixed hostname processing in hostlist file so hostnames are always converted to lower case so they will match what's returned in the PE attach.cfg file.

Comment 4 Beth Tibbitts

2011-06-13 14:07:51 EDT

Problem is also in PTP 5.0 - assuming we're fixing in in both places?

Comment 5 David Wootton

2011-06-13 14:31:18 EDT

Yes, this is fixed in both 4.0 and 5.0

Comment 6 David Wootton

2011-06-13 15:42:41 EDT

Fixed in 4.0 and head