Skip to main content
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] [List Home]
Re: [linuxtools-dev] TMF 
Many thanks for all your answers.
It's seems CTF/TMF  is definitively the best solution.
I've just another question: Where can I found documentation about TMF model ? 

Regards,

Xavier Raynaud


On 01/24/2012 08:32 PM, Aaron Spear wrote:

Hello Xavier,

As Dominique said, it has been the ambition of the Multi-core
associations tool infrastructure working group to extend CTF to cover
precisely the use case you mention among others. We (I am the co-chair
of the group) have spent a lot of time thinking about how CTF can cover
various multi-core tracing use cases with correlation of different trace
streams, handling different clock domains, and all of the other
associated issues. Dominique pointed to a paper I wrote in one of the
links below that talks about the ambitions.

The Ericsson guys on this thread (Dominique, Francois) and others
involved with the Linux tools LTTng viewer were forward thinking with
their design of TMF and the view components on top of it, so it is a
good architecture to scale to other platforms, there is just work to do
since they started with Linux (perfect for you to jump in and
contribute!). The funny thing is that at my previous employer (Mentor
Graphics) I had designed the data model that I wanted for a general
multi-core system analyzer and in the end someone pointed me to TMF and
I found that the model was almost exactly what I designed independently.
(So, Mentor's system analyzer product
<http://www.mentor.com/embedded-software/sourcery-tools/system-analyzer>
does build on some of this code, though I don't know the current state
or their plans).

It has been my hope that the Linux tools LTTng analyzer can become a
useful platform on which heterogenous multi-core analyzers can be built.
That in one product you would be able to see Linux system traces
aggregated with hardware trace of other cores, instrumentation trace for
proprietary kernels or "roll your own" bare metal apps, and perhaps most
interestingly also allowing understanding of application behavior in the
context of the many layers of software that we increasingly find as we
march towards "the cloud".

regards,
Aaron Spear
MCA Tools Infrastructure working group co-chair

------------------------------------------------------------------------

    *From: *"Dominique Toupin" <dominique.toupin@xxxxxxxxxxxx>
    *To: *"xavier raynaud" <xavier.raynaud@xxxxxxxxx>, "Linux Tools
    developer discussions" <linuxtools-dev@xxxxxxxxxxx>
    *Sent: *Monday, January 23, 2012 9:00:20 AM
    *Subject: *Re: [linuxtools-dev] TMF

    CTF is not Linux specific, the multicore association tool
    infrastructure work group was involve to define the requirement/spec.

    Yes LTTng/Linux/Eclipse is the only open source implementation so
    fare but the good news is if you are using CTF for other use cases
    e.g. HW tracing, you will be able to re-use many cmpts, e.g. parser
    (both Java and C/C++) and many views/correlation in Eclipse

    http://www.multicore-association.org/workgroup/tiwg.php

    http://www.efficios.com/ctf

    http://www.eetimes.com/design/embedded/4214860/Using-trace-to-solve-the-multi-core-system-debug-problem

    http://www.multicoredevcon.com/common/session.php?expo_seq=12&track_seq=159&pres_seq=958
    <http://www.multicoredevcon.com/common/session.php?expo_seq=12&track_seq=159&pres_seq=958>

    http://www.eclipsecon.org/europe2011/sessions/efficient-cc-tracing-eclipse

    *From:*linuxtools-dev-bounces@xxxxxxxxxxx
    [mailto:linuxtools-dev-bounces@xxxxxxxxxxx] *On Behalf Of *Francois
    Chouinard
    *Sent:* January-23-12 9:46 AM
    *To:* Linux Tools developer discussions
    *Subject:* Re: [linuxtools-dev] TMF

    Yes it is. CTF is an efficient trace format (based on research done
    for LTTng) that can be used for essentially anything (kernel,
    user-space, hardware, ...)

    We will also provide a CTF parser with Juno (we have a few wrinkle
    to iron out but it is practically ready for a CQ). It should appear
    in HEAD (well, 'master') shortly.

    Although TMF and LTTng will have a dependency on it, this CTF parser
    generator will be a component on its own that can be re-used without
    the rest (event without Eclipse...)

    Regards,

    /fc

    On Mon, Jan 23, 2012 at 9:28 AM, Xavier Raynaud
    <xavier.raynaud@xxxxxxxxx <mailto:xavier.raynaud@xxxxxxxxx>> wrote:

    Hi,

    Excellent.
    I've read somewhere that one CTF objective is to be system-agnostic
    (even if, for now, focused only on linux). Is it still true ?

    Many thanks,
    Xavier





    On 01/23/2012 03:21 PM, Francois Chouinard wrote:

        Hi,

        You can start by looking at TMF itself. Some if not most of the
        LTTng
        views/widgets are really generic TMF views/widgets that have been
        extended for LTTng purposes.

        For the context-switching et al., we initially simply ported the
        GTK-based LTTV's State System (the component that handles the
        processes/resources state transitions based on the events sequence).
        This State System is really Linux kernel-oriented.

        We are now working on a persistent generic State System (a part
        of TMF
        itself) that will be re-usable for any type of state management. It
        should be integrated in the coming months and delivered with
        Juno. You
        might want to consider basing your work on this. Constraint:
        Although
        the new State System will be trace-format agnostic, for the initial
        release we are likely to focus on CTF (Common Trace Format) traces.

        Don't hesitate to contact us for more details.

        Regards,
        /fc


        On Mon, Jan 23, 2012 at 3:39 AM, Xavier Raynaud

        <xavier.raynaud@xxxxxxxxx <mailto:xavier.raynaud@xxxxxxxxx>
        <mailto:xavier.raynaud@xxxxxxxxx
        <mailto:xavier.raynaud@xxxxxxxxx>>> wrote:

        Hi,

        In the coming weeks, I will start to design a GUI to display traces
        for a massively parallel device.

        My first idea was to use TMF for that - and do something similar to
        the LTT-ng plugin.

        For now, this device does not run linux, but the traced event will
        be very similar (context-switches, interrupts, user events...)

        Is there any hint available somewhere, or any trap to avoid ?

        Many thanks,

        Xavier Raynaud

        _________________________________________________
        linuxtools-dev mailing list
        linuxtools-dev@xxxxxxxxxxx <mailto:linuxtools-dev@xxxxxxxxxxx>
        <mailto:linuxtools-dev@xxxxxxxxxxx
        <mailto:linuxtools-dev@xxxxxxxxxxx>>
        https://dev.eclipse.org/__mailman/listinfo/linuxtools-__dev
        <https://dev.eclipse.org/mailman/listinfo/linuxtools-dev>




        --
        Francois


    _______________________________________________
    linuxtools-dev mailing list
    linuxtools-dev@xxxxxxxxxxx <mailto:linuxtools-dev@xxxxxxxxxxx>
    https://dev.eclipse.org/mailman/listinfo/linuxtools-dev



    --
    Francois


    _______________________________________________
    linuxtools-dev mailing list
    linuxtools-dev@xxxxxxxxxxx
    https://dev.eclipse.org/mailman/listinfo/linuxtools-dev

Back to the top