Messages in this thread | | | From | "Liang, Kan" <> | Subject | RE: [PATCH RFC V2 3/5] perf,tool: partial time support | Date | Mon, 13 Jul 2015 18:59:17 +0000 |
| |
> -----Original Message----- > From: Jiri Olsa [mailto:jolsa@redhat.com] > Sent: Sunday, July 12, 2015 9:22 AM > To: Liang, Kan > Cc: acme@kernel.org; jolsa@kernel.org; namhyung@kernel.org; > ak@linux.intel.com; linux-kernel@vger.kernel.org > Subject: Re: [PATCH RFC V2 3/5] perf,tool: partial time support > > On Tue, Jul 07, 2015 at 06:19:03AM -0400, kan.liang@intel.com wrote: > > From: Kan Liang <kan.liang@intel.com> > > > > When multiple events are sampled it may not be needed to collect fine > > grained time stamps on all events. The sample sites are usually nearby. > > It's enough to have time stamps on the regular reference events. > > This patchkit adds the ability to turn off time stamps per event. This > > in term can reduce sampling overhead and the size of the perf.data. > > how is this usable in some perf example? I dont get the correlation > between 'refference' and the rest of the events in the report time.. >
One usable case of partial time is to work with partial callgraph to enable "PEBS threshold > 1" (https://lkml.org/lkml/2015/5/10/196), which can significantly reduce the sampling overhead. For the case as above, we intend to get callgraph in large PEBS threshold. Timestamp is not that important. So I didn't correlate the reference event and the rest events here.
But it's doable. We can use oe->last_flush to set the rest events' time. The only problem is that I cannot find a useable partial time only case. (Partial time itself cannot save as much as partial callgraph) So I think we do not need to implement such correlation.
I guess I will keep the code unchanged, but correct the description. Is it OK?
Thanks, Kan
| |