On 23/04/15 11:23, Mark Goodwin wrote:
Hi Ken, just trying to review these changes ...
Thanks.
... - I guess the crux
of the issue falls on the "fixed logic errors elsewhere", as mentioned :
...
These changes are in the previous half dozen or so commits involve interp.c
I built and installed with these changes and checked qa/787 passes.
But I've lost Frank and Paul's recipe for reproducing the archive replay
performance issue - anyone have a reference to that handy?
qa/787 uses qa/archives/bug-1044 which is a pmlogextract'd version of
Frank's original super-SLOW-merged-archive-20140124.225919 archive. A
tarball of the original archive is on oss.sgi.com:~kenj if you need/want
it (~kenj/super-slow-archive.tar.xz). With this archive unpacked:
$ time pmval -t 60 -a super-SLOW-merged-archive-20140124.225919
pmcd.numagents
should finish in seconds, not minutes!
Paul's one is trickier ... I don't have an automated test for that and
I've been testing interactively also with pmlogextract'd subset of the
metrics ... gimme a while to snoop the details out.
|