Re: Problem with Itanium-I Server and numerical programs

[David Mosberger <davidm@napali.hpl.hp.com>]

>>>>> On Tue, 11 Jan 2005 18:23:21 +0100, Wolfram Ruehaak <w.ruehaak@gga-hannover.de> said:

  Wolfram> Since some time (don't know exactly since when) large
  Wolfram> calculations (finite-element and finite-difference) hang or
  Wolfram> freeze: after some calcualtions (iterations and/or
  Wolfram> time-steps) there is absolutely no more progress, the job
  Wolfram> just hangs, but with > 99 %CPU.  I have waited for weeks,
  Wolfram> but there is no progress.

Have you tried attaching to the running process with gdb (and/or strace)?
In gdb, you'd do this with:

	gdb PROGRAM -p PID

if you single-step or do some manual IP sampling, it should at least
give you some idea where it's spinning and from there, you might get
some ideas where to look next.

Does the problem occur _only_ on Merced machines or McKinley/Madison, too?

	--david