Problems with the SP/2

Core files

When AIX generates a core dump, it doesn't dump the data segment of the program to the core file. The stack is dumped, which is usually enough for a stack trace, so you can see where the program failed.

However, that's not good enough for Shore, as the stacks for all the threads are in the data segment. The good news about this is that I have a uncommited fix which will cause the data segment to be dumped. The bad news about this is that gdb doesn't debug core files yet, so it doesn't really matter.

Gdb

The current gdb distribution doesn't work on the SP/2. I have my own version, in ~bolo/bin/gdb. It may be of use to you, or it may fail. Here are some known problems:
  1. It will not debug a core dump.
  2. It may not debug some live executables. Gdb will dump core when loading symbols from the exectutable.
  3. It prints debugging messages that only bolo understands.

/etc/group

The SMIT SP/2 administration tools don't update the contents of /etc/group on the nodes. To be more precise, node-01 may be updated, but none of the other nodes are. I have shell script in ~bolo/SP2/update_group which will distribute changes to /etc/group.

See Adding users to the SP/2 for more information.

Poe

The Parallel Operating Environment needs exclusive use of at least the following resources: This breaks any Unix programs which utilize these facilities.

I am modifying shore to work around these problems. In the interim, either use pexec to start jobs on the nodes, or create some xterms and start jobs by hand.


SP/2 Information
Bolo's Home Page
Last Modified: Mon Dec 18 13:22:50 CST 1995
bolo (Josef Burger) <bolo@cs.wisc.edu>