This one is totally CPU bound ... it used to run for 220-odd seconds on
my clunker workstation (I've no idea what Nathan's using to stretch that
to 800+ seconds), and I can get it down to 70-odd seconds with a couple
of quick optimizations ... but it is exploring 4096 cases (considering
each word-aligned truncation of the last 16K bytes of the archive to
mimic what might happen with aborted stdio buffering of the output from
pmlogger) for each of 3 archives, and without remembering exactly what
triggered the problem in the past, I'm not sure I can make it do less
work without changing the test coverage.
On Wed, 2009-03-04 at 15:00 +1100, Nathan Scott wrote:
> On Wed, 2009-03-04 at 14:55 +1100, Max Matveev wrote:
> > On Wed, 04 Mar 2009 12:36:20 +1100, Nathan Scott wrote:
> >
> > nscott> Thats well over 10 minutes. Its a "pmdanasavg SEGVs in
> > nscott> pmGetArchiveEnd" test with owner=kenmcd. :)
> >
> > pmdanasavg was since re-written to use HUTA mode and not do archive
> > tailing. The test is still sort of useful.
>
> Yep, no disagreement that it looks useful ... it just takes far
> too long. There must be a way to cut it down to, say, a minute
> or less like most other tests.
>
> cheers.
>
> --
> Nathan
>
|