Hi Stavros,
First, thanks for the interest :)
To add support for a new trace type to TMF, you basically have 2
options: either you first convert your trace to another supported
format, or you write a parser in TMF. In practice, the second option is
easier. There is, imho, not much gain in converting existing traces to
CTF, that just ends up being an extra intermediate format. (On the other
hand, if you tracer directly writes in CTF, that can net you advantages).
If your trace is in text format, it's possible to create a custom text
parser, which you can define directly from the UI using regular
expressions. If it's a binary format, then that requires writing some
Java code.
Next, you'll typically want to add views/analyses for your trace type.
The view in your screenshot look quite doable, using our generic Time
Graph View. You can see a screenshot of the Control Flow and Resource
View, both based on the Time Graph views, on that page:
http://eclipse.org/linuxtools/projectPages/lttng/
(Latest version also supports showing arrows in the time graph. The
latest Control Flow View does so, to show the "flow" of each CPU. We
should update that screenshot...)
Another thing to note, a very new feature in TMF is the possibility to
define analyses and views using XML definitions, so you do not need to
write Java code to create a custom view. This makes it VERY easy to
create new views! The data-driven views are a bit more limited than the
"Java-driven" ones (for example, I think right now arrows are not
supported in the data-driven views), but it's slowly improving.
If you have any other questions, please let us know!
Cheers,
Alexandre
On 06/12/2014 07:56 AM, Stavros Giannakopoulos wrote:
Greetings,
i am running a project on behalf or Ericsson Sweden regarding tracing on a
custom multicore architecture.
I came across your tools through a colleague and It seems that they might be
appropriate for our work, especially TMF since we already utilize
eclipse quite heavily.
I have some technical questions though, and I would like some feedback so I
know whether our specifications can be covered by that or some similar tool.
Please forward this email to the appropriate department if possible.
The specifications we have are:
1. The trace we are interested in is a process-centric view of the events (or
actors in our system) along with their respective threads, that will run on
specific cores.
2. We don't need to capture the traces, that has been already done, only to
parse the generated file and produce a graphical view that includes the
processes, threads, cores and the messages that trigger new processes.
is that possible?
3. We have the trace files in .lat format already which are generated by using
the Flexus tools. What I want to know if there is a conversion tool available
or we need to develop a script to do the conversion to a suitable format (CTF).
And if the TMF or LTTV can produce this result from a custom trace file and not
a kernel trace.
Observe on Fig 1 a simplified view of our needs. In practice there will be a
few thousands processes in tens of cores and the start/end times and the
messages are essential.
Fig 1: Example of result wanted.
[cid:[email protected]]
Please reply on my email:
[email protected]<mailto:[email protected]>
or contact Johan Mellgren at
[email protected]<mailto:[email protected]>
Regards,
Stavros Giannakopoulos
_______________________________________________
lttng-dev mailing list
[email protected]
http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
_______________________________________________
lttng-dev mailing list
[email protected]
http://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev
