Messages in this thread | | | Date | Thu, 24 Jul 2014 17:20:56 +0200 | From | Jiri Olsa <> | Subject | Re: [PATCHv3 00/19] perf tools: Factor ordered samples queue |
| |
On Thu, Jul 24, 2014 at 12:10:09PM -0300, Arnaldo Carvalho de Melo wrote:
SNIP
> > > > should be good to go.. > > > > Namhyung pointed out performance implications for patch 17: > > perf tools: Always force PERF_RECORD_FINISHED_ROUND event > > I haven't seen this comment, just saw David, IIUC, having problems using > the whole patchkit, David? > > My tests were more to counter what he said, i.e. that: > > <quote David Ahern> > No overload condition, no dropped chunks message - yet can't process > events > </> > > I see no overload condition, no dropped chunks message, I was pounding > the machine with a big load and could process the events, so all seems > well. > > And from the discussion I understood that patch 17 is the most important > one, agreed by you and Adrian, right?
yes, it drives the flush (in report) to happen
Namhyung pointed out, that now we store extra event for non-tracepoint events sessions.. which might have performance implications:
http://marc.info/?l=linux-kernel&m=140266032913387&w=2
> > I.e. the other ones are refactorings to make it have better naming and > to allow it being used elsewhere (builtin-trace.c, whee), but still need > some more polishing/testing, is that right?
correct.. I plan to polish the debug output patchset, I dont think I've got comments for the rest
jirka
| |