----- Original Message -----
> On Fri, 2012-12-14 at 01:53 -0800, Jun Wang wrote:
> > In terms of the add-hoc high frequency stats/metrics collection,
> ...
> The trace PMDA was never intended for high frequency sampling
> ...
On the topic of pmdatrace I would also add that this PMDA pre-dates
many modern advances in event tracing - in particular, concepts like
end-to-end tracing (implemented in Event Tracing for Windows, a number
of commercial products like DynaTrace, New Relic, and so on), dynamic
tracing (systemtap, DTrace, dyninst and co), low-cost static tracing
(like LTTng), and ironically even the recent event tracing extensions
in PCP itself.
pmdatrace is well overdue for a revamp, so unless your needs are quite
simple I wouldn't really recommend using it at this stage.
> More interesting would be the relatively recent PMAPI extensions for
> event records ... Nathan can probably expand on this better than I.
Further details can be found in the paper here, from page 11 onward:
http://oss.sgi.com/projects/pcp/papers/unifying-thesis.pdf
This is relatively fresh code in PCP, and so not as well documented
at this stage as other areas. In general, this is an area of active
exploration as we're uncovering the needs of each new tracing domain
we come across.
cheers.
--
Nathan
|