Date: Thu, 30 Oct 2014 22:26:21 -0700 From: Brendan Gregg <brendan.d.gregg@gmail.com> To: Ed Maste <emaste@freebsd.org> Cc: Stefan Parvu <sparvu@systemdatarecorder.org>, freebsd-performance@freebsd.org Subject: Re: pmcstat -z32 -G truncates callgraph to 8 Message-ID: <CAE40pdeQvmEDQRZo=v06Dzg7n0%2BgHNqPNvWp3R7cAQ2bBd3yoA@mail.gmail.com> In-Reply-To: <CAPyFy2Df6NO4x-NG7eUiCbN8_9%2BvNAdtSnTg%2B%2BPSnnoYeiEStg@mail.gmail.com> References: <CAE40pdf5AzR4tuGZUQugoK6jfbvd_dLLBtdfv2FPQYY3Lb5PrA@mail.gmail.com> <20141028114537.83f9f0a3c730949f453c3819@systemdatarecorder.org> <CAE40pdf6jUGEse0Mh%2BgV48VnwLrWS%2BfhO59gfzY=ocRauK43MA@mail.gmail.com> <CAPyFy2Df6NO4x-NG7eUiCbN8_9%2BvNAdtSnTg%2B%2BPSnnoYeiEStg@mail.gmail.com>
next in thread | previous in thread | raw e-mail | index | archive | help
G'Day Ed, On Wed, Oct 29, 2014 at 12:40 PM, Ed Maste <emaste@freebsd.org> wrote: > On 28 October 2014 13:38, Brendan Gregg <brendan.d.gregg@gmail.com> wrote: >> Ah, thanks, I'm on 10.0-STABLE and I have: >> >> kern.hwpmc.callchaindepth: 8 >> >> Glad it's something simple! > > Those are the best kinds of problems to have, although we ought to > make sure this point is covered in the FreeBSD profiling documentation > that we have or create. Yes; at least we have this thread now, which should be searchable. > > Also, do you think that we should bump the compiled-in default up to 32? Yes. When I'm using profiling data, I like full stacks for making flame graphs. For the FreeBSD kernel, 32 frames should usually be enough (I have a flame graph that reaches 24 frames for the kernel, but no more). For user-level, I'd probably need ~100. So making the compiled-in default to 32 would hopefully be sufficient for most kernel profiling, and one would need to bump that up for deep user-level stacks. I guess this would also need PMC_CALLCHAIN_DEPTH_MAX = 128 to work. Brendan
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?CAE40pdeQvmEDQRZo=v06Dzg7n0%2BgHNqPNvWp3R7cAQ2bBd3yoA>