GSoC Project | Basic support for Trace Compass

[Ravindra Kumar Meena]

>
> I don’t really get why you define 32 different streams in your metadata
> which means you have to copy the event definition as well 32 times.
>
I added 32 different streams in metadata because I have 32 binary stream
files stored in the disk. I have defined one type of event for each stream
file.
e.g.
stream{
id = 0;
};

has one event type:

event {
  name = "RTEMS_RECORDING_EVENT_CPU_0";
  stream_id = 0;
  fields := struct event_field;
};
similarly,

stream{
id = 1;
};

has one event type:

event {
  name = "RTEMS_RECORDING_EVENT_CPU_1";
  stream_id = 1;
  fields := struct event_field;
};

and so on.......

This amount of duplication seems inflexible and with a lot of overhead, so
> I would assume there is a more elegant way.
>
>
>
> If I understand the documentation correctly, then you can have multiple
> files containing data of a single stream (type). You just need the proper
> headers at the beginning of each file.
>
I added <stream_id> at the beginning of every stream files.
So final binary stream format looks like this:
<stream_id> <ns> <event> <data> <ns> <event> <data> .....

> If you have one trace file per CPU, which all generate the same types of
> events and only differ regarding the respective cpu id, isn’t something
> like in the examples of the CTF documentation possible (
> https://diamon.org/ctf/#ex-packet-context) ?
>
> Then, each file would contain one “ctf packet” where the packet header
> defines the cpu id and then a consecutive list of the events of this cpu.
>
I did it once. Is this something you are suggesting?
https://github.com/rmeena840/rtems-tools/commit/2f9e7b49f55ccc39613eb277fa410eb9764fab1d


>
>
> Also, in your previous metadata description it looks like you already
> encode the cpu id as part of the event header, so why then the duplication?
> And why was the cpu id in the event header, shouldn’t it be in the event
> context?
>
Okay. I have to work on the header part of the binary stream files.

>
>
> Sebastian mentioned a more complex format necessary for the TraceCompass
> application. I have not checked this.
>

-- 
*Ravindra Kumar Meena*,
B. Tech. Computer Science and Engineering,
Indian Institute of Technology (Indian School of Mines)
<https://www.iitism.ac.in/>, Dhanbad
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/devel/attachments/20190710/d437cffd/attachment-0002.html>