Bug 953759

Description of problem:

PCP pmlogconf entries needs a bit of a cleanup. This is proposed as
part of deriving a generally useful pmlogger default config that
will suit most sites out-of-the-box, thus improving useability and
reducing the learning curve for new users.

> ...
> make sense. I'll make a start on this approach, beginning with
> spring cleaning the existing stuff; there are some very IRIXY
> entries that should be just nuked (kernel/*-irix). And some that
> need Linux porting (e.g. filesystem/xfs-io-irix). Others need

*nod*

> reviewing and I think we need a whole new group based on the
> metrics required by various tools (pmclient, pmstat, sar, iostat,
> vmstat, collectl / pmcollectl and so forth).

Good idea.  Rather than keeping two copies of the config files for
each tool, it'd be worthwhile moving the existing configuration
(find src -name pmlogger.config) into pmlogconf/tools and using it
as a starting point.  Will clean up /etc/pcp/pmlogger too without
all those in there - bonus.

> The resulting set of metrics (as a more-or-less configurable union)
> should constitute a reasonable default logging config - allowing most
> tools to replay archives, with enough extra scope for drill-down with
> generic tools (pmchart esp.). We'll also need to measure daily data
> volumes on variopus sized machines and adjust sampling rates so other
> tools like sos-report wont balk, and also so the data collection
> itself remains reasonably unintrusive on system resources.

*nod* ... maybe add -r option to pmlogger in the control file too?

Additional info:

Currently shipping /etc/pcp/pmlogger/config.default is way too terse :

log mandatory on once { hinv.ncpu hinv.ndisk }

log mandatory on 1 hour
    kernel.all.load [ "1 minute" ]