[ltt-dev] Multi-second gaps in trace events

Mathieu Desnoyers compudj at krystal.dyndns.org
Wed Sep 30 15:09:40 EDT 2009


* Chad Willman (cwillman at lifesize.com) wrote:
> 
> No dmesg warnings about either of these.
> 
> I haven't done anything specific with the buffer size. Let me know if I just
> need to RTFM.
> 
> I'm running in "flight" mode, which I believe is overwrite.
>

Ah! this is it then. You run in flight recorder mode. So when the
buffers are filled quicker than lttd can read them, the oldest buffers
are overwritten, which creates "gaps" in the timeline.

Flight recorder mode is more made to :
- trace
- hit interesting problem
- stop tracing
- grab a snapshot of past activity by running lttd.

If you want to take a trace of your system activity between trace start
and trace stop, then you need the "normal" mode.

You will probably need to increase the size of your buffers with lttctl
though, because if lttng overwrites in flight recorder mode, you will
very likely have "events lost" in normal mode.

Mathieu
 
> Chad
> 
> 
> On 9/30/09 12:44 PM, "Mathieu Desnoyers" <compudj at krystal.dyndns.org> wrote:
> 
> > OK. Then, other question, do you have warnings in your traced board
> > dmesg telling that you have "lost subbuffers" or lost events ? This is
> > shown right after tracing stops. (newer LTTng has this warning shown in
> > LTTV, but your old LTTng/LTTV combo doesn't)
> > 
> > Your buffer size may be too small.
> > 
> > Are you tracing in "overwrite" or "non-overwrite" mode ?
> > 
> > Mathieu
> > 
> > * Chad Willman (cwillman at lifesize.com) wrote:
> >> 
> >> Thanks for the reply.
> >> 
> >> Yes, it's an ARM, but it has a cycle-counter and we've actually added
> >> support for it to LTTng. We get nanosecond resolution.
> >> 
> >> Chad 
> >> 
> >> 
> >> On 9/30/09 12:29 PM, "Mathieu Desnoyers" <compudj at krystal.dyndns.org> wrote:
> >> 
> >>> * Chad Willman (cwillman at lifesize.com) wrote:
> >>>> 
> >>>> I¹m seeing long (multi-second) gaps in trace events and am wondering if
> >>>> this
> >>>> is a known issue.
> >>>> 
> >>>> For example, trace events shown in the viewer appear normal and then
> >>>> suddenly there will be a gap of 9 seconds (varies) where it appears that a
> >>>> single thread is running in user mode (green) and nothing else is happening
> >>>> on the entire system in any thread for 9 seconds (no context switches, . .
> >>>> .
> >>>> nothing).
> >>>> 
> >>>> This is an excellent tracing facility, but I can¹t rely on it entirely
> >>>> until
> >>>> I have no gaps in the trace. Any ideas ?
> >>>> 
> >>>> I¹m running -
> >>>> Lttng v 0.6.33 on Linux 2.6.18 kernel on a TI DaVinci DM6467T SOC platform
> >>> 
> >>> That's an ARM board right ?
> >>> 
> >>> On these boards, LTTng has a 1 jiffy precision, because it uses the
> >>> "generic clock". Within this interval, events are simply counted, e.g. 1
> >>> cycle = 1 event.
> >>> 
> >>> The ARM OMAP3 has cycle counters, which LTTng uses.
> >>> 
> >>> Mathieu
> >>> 
> >>>> Ltt-control v 0.30
> >>>> Lttv v 0.8.72
> >>>> Genevent v 0.30
> >>>> 
> >>>> Regards,
> >>>> 
> >>>> CW
> >>> 
> >>>> _______________________________________________
> >>>> ltt-dev mailing list
> >>>> ltt-dev at lists.casi.polymtl.ca
> >>>> http://lists.casi.polymtl.ca/cgi-bin/mailman/listinfo/ltt-dev
> >>> 
> >> 
> 

-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68




More information about the lttng-dev mailing list