[
Date Prev][
Date Next][
Thread Prev][
Thread Next][
Date Index][
Thread Index]
[
List Home]
[tracecompass-dev] Visualize trace analyses for experiments
|
Hello
I was investigation how to visualize trace analyses when the trace is
part of an experiment. Right now, there are different behavior depending
on the type of view.
For Time Graph views like Control Flow view, Resources view, Call Stack
view all LTTng kernel traces are shown in the same view. The entries
(rows) are grouped by traces. The XML time graph view is designed the
same way, however, it's seems to be broken currently.
For XY chart views, like CPU usage view, memory usage usage etc. now
trace data is shown if they are in an experiment. Not sure yet about the
XML XY chart view, but it's probably the same
Other views, like the System Call table or Stream List view (pcap
traces) won't be populated either for experiments.
The Statistics view and Histogram view show an aggregation of the data.
The approach for Time Graph views where the data is grouped by trace is
acceptable and can stay.
In the following I'd like to discuss a view options to visualize trace
analysis when part of an experiment for the other types of views:
1) Show the trace data of one selected trace of the relevant trace type
1.1) The view could select the first trace of the relevant type. This
this would not cover the case if there are multiple traces of the same
trace type that could fill the view.
1.2) Add a view widget to select the trace to be used for populating the
view. This could be drop-down menu in the toolbar where the user can
change the trace of interest.
2) Visualize the data of all traces of the relevant trace type
This approach could be take for some of the views where it is possible
to group the data per trace. For example, in a XY line chart, each line
can represent the data of one trace. For tables, like the System Call
Table, the data could be grouped by traces (similar to the way it is
done in the Time Graph views). However, for views showing density, this
approach might not be the best, because user would like to see the
density per trace.
Right now I'm leaning to option 1.2). If needed, option 2 can be
implemented afterwards and it could be a separate entry in the drop down
menu.
Any thoughts?
Cheers
Bernd