[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
|
Re: [tracecompass-dev] Synchronizing LTTng Kernel and UST Traces ?
|
Hi Matthew and Francis,
Many thanks for considering this. The trace is attached on
https://bugs.eclipse.org/bugs/show_bug.cgi?id=484620
I was not 100% sure if "lttng" commands are correct for how my trace was configured, since I had done that partially through the UI; but wanted to give you at least an idea of what's expected to be in the trace.
The trace is entirely taken from inside the QEMU, so the QEMU acts like a real host here (although a slow one with only one CPU and unfsd-mounted rootfs). So perhaps the system was just too loaded to provide reasonable data ...
Thanks,
Martin
--
Martin Oberhuber, SMTS / Product Owner - Development Tools, Wind River
direct +43.662.457915.85 fax +43.662.457915.6
-----Original Message-----
From: tracecompass-dev-bounces@xxxxxxxxxxx [mailto:tracecompass-dev-bounces@xxxxxxxxxxx] On Behalf Of Matthew Khouzam
Sent: Thursday, December 17, 2015 6:04 PM
To: tracecompass-dev@xxxxxxxxxxx
Subject: Re: [tracecompass-dev] Synchronizing LTTng Kernel and UST Traces ?
Hi, I would like to clarify this a bit, and I also suspect that this may be more of an LTTng issue.
Is the kernel trace of the host (virtualisor) or the target (virtualisee)?
If your traces are not too large, it would be very cool to see them in a bug. We can also in that case, if it is a bug, use them as a regression test. (if you're ok with them being in our test suites)
Thanks for the continued implication in the project, we really appreciate these questions,
Matthew
On 15-12-17 03:52 AM, Oberhuber, Martin wrote:
>
> Hi Tracecompass Experts,
>
>
>
> I have obtained an LTTng-UST and Kernel Trace in parallel, but looking
> at the respective combined experiment it turns out that Kernel and UST
> seem to use different time sources (there's a time skew). Consider
> attached screenshot: The UST event is shown in the middle of syscall
> state, but it would be expected in "lemon_server RUNNING" state.
>
>
>
> Reading the Trace Compass docs, I see that there's a synchronization
> feature, but only for network traces and only with experimental
> branches enabled.
>
>
>
> What could be done to get Kernel and UST traces synchronized ? Or
> should a completely different method of data gathering be considered,
> eg doing "perf record" with uprobes enabled and then converting the
> perf.data to CTF for investigation ? Correlating application traces
> with Kernel event seems a powerful tool to me for understanding
> excessive latencies . perhaps adding UST pthread events could help
> with trace synchronization since these should exist on Kernel and UST
> side ?
>
>
>
> Many thanks for any pointers !!
>
>
>
> PS my sample trace was obtained on Linux qemux86, Kernel 4.1 with
> lttng-2.6. I'm happy to share the recorded traces if anybody is
> interested.
>
>
>
> Thanks,
>
> Martin
>
> --
>
> *Martin Oberhuber*, SMTS / Product Owner - Development Tools, *Wind
> River*
>
> direct +43.662.457915.85 fax +43.662.457915.6
>
>
>
>
>
> _______________________________________________
> tracecompass-dev mailing list
> tracecompass-dev@xxxxxxxxxxx
> To change your delivery options, retrieve your password, or
> unsubscribe from this list, visit
> https://dev.eclipse.org/mailman/listinfo/tracecompass-dev
_______________________________________________
tracecompass-dev mailing list
tracecompass-dev@xxxxxxxxxxx
To change your delivery options, retrieve your password, or unsubscribe from this list, visit https://dev.eclipse.org/mailman/listinfo/tracecompass-dev
