Hi Nathan,
I have made some slight changes to pmdagfs2 with regards to efficiency
and some extra filtering of glocks w.r.t gfs2_glock_lock_time metrics.
Most of the insertions and deletions are due to moving functions around
in the source to negate the need for the forward references at the
beginning of lock_time.c
Sorry for leaving this a little late to commit, was waiting on some testing
on a real cluster (still on-going but looking promising) in addition to my
tests on my virtual cluster.
Please let me know if there are any issues.
Cheers,
Paul.
Changes committed to git://github.com/pauljevans/pcp.git dev
commit a3f9ecfe1de9b75552d6f8f321865aa626184078
Author: Paul Evans <pevans@xxxxxxxxxx>
Date: Tue Jul 30 16:57:41 2013 +0100
pmdagfs2: Additional filtering and efficiency improvements
Improved filtering for glock_lock_time metrics by glocks now requiring
a minimum threshold (currently 25 can be increased) for the dlm and
queue fields ensuring that the locks chosen have been through the dlm
the given number of time. We filter only for locks that have been
passed
around a given number of times.
Made changes to the sorting and comparison functions allowing a
simpler
choice w.r.t the worst glock at the given time (also easier to read).
These changes are in place ready for future improvements (top 10
glocks etc).
Make some slight changes in the comments to reflect the changes in the
code and work towards reducing redundant comments.
src/pmdas/gfs2/lock_time.c | 252
+++++++++++++++++++++------------------------
src/pmdas/gfs2/lock_time.h | 21 ++--
2 files changed, 132 insertions(+), 141 deletions(-)
|