Re: [tracecompass-dev] [Question] Using Trace Compass with Custom Schedu

[Matthew Khouzam <matthew.khouzam@xxxxxxxxxxxx>]

Additional Questions:

1. Are there any existing examples of using Trace Compass with custom RTOS/schedulers?

Linux analyses are a great example. However, there are others, such as https://link.springer.com/article/10.1007/s11219-016-9311-0 if academic research is OK. It reads windows traces!

2. What are the minimum trace data points needed to create a meaningful visualization?

If you look at our test cases, sched_switch is the minimum necessary. I like having sched_wakeup, sched_ttwu, irq* and system calls too. But minimum is sched_switch.

3. Are there any performance considerations when implementing tracing on an embedded target?

Many, if there weren't, it would be a solved problem.  1- You have limited memory, 2- you have limited IO, 3- you have limited CPU, 4- you have limited storage.

So I would recommend logging little and logging compressed unconverted data then letting the trace viewer do the heavy lifting.

Hope this helps.

---

From: tracecompass-dev <tracecompass-dev-bounces@xxxxxxxxxxx> on behalf of Matthew Khouzam via tracecompass-dev <tracecompass-dev@xxxxxxxxxxx>
Sent: Thursday, October 10, 2024 1:48 PM
To: tracecompass-dev@xxxxxxxxxxx <tracecompass-dev@xxxxxxxxxxx>
Cc: Matthew Khouzam <matthew.khouzam@xxxxxxxxxxxx>
Subject: Re: [tracecompass-dev] [Question] Using Trace Compass with Custom Scheduler on TMS570LS1114 MCU

 

Hi Ahmed,

I will give personal opinions with some years of experience backing them, please treat them as such.

1. What is the recommended approach for:
• Collecting trace data from a custom scheduler on an embedded target

MATTHEW: I see a few ways to do this. I would base myself on pre-existing solutions to save headaches/time/money/sanity.

Trace event Traces (google chrome traces): they are json, easy to read as a human, they give decent results. You will be able to visualize the scheduler on a flame chart. But they are not specialized in embedded. For that, you can look into....

BTF: you can make a btf compliant trace, it would be good if you want to also trace actuators, SPI busses or something like that.

I find it good for simple MCUs, but when you go multi-core, I really prefer.... 

Linux style traces. This is my personal favorite because it gives more liberty. If you have a sched_switch event going from last to current, you can measure stuff like time to wakeup, time to schedule and more. 

• Converting the collected data into a format that Trace Compass can understand

The CTF format is very embedded friendly as it requires little computation. Look at barectf(https://github.com/efficios/barectf) as a writer to help out on this side. I think   normand (https://github.com/efficios/normand) can help.

• Visualizing task switches and execution times

Depending on your approach you can use the control flow view, resources view and other like flame graph. I am biased, I think the latest trace compass version is the greatest. However, https://www.ti.com/tool/CCSTUDIO if you use code composer studio, it is based on eclipse IDE or Theia IDE, and you can integrate trace compass into it.

2. Specifically, I need to know:
• What trace data format would be most suitable for this use case

If you want something long term, I suggest CTF is by far the best and most efficient. If it's a weekend project, I would suggest looking at BTF. No the formats are NOT related.

• How to implement the necessary trace points in the scheduler

I would inspire myself on the linux kernel's way of doing it. If you look here:https://github.com/eclipse-tracecompass/org.eclipse.tracecompass/blob/master/analysis/org.eclipse.tracecompass.analysis.os.linux.core/src/org/eclipse/tracecompass/analysis/os/linux/core/trace/IKernelAnalysisEventLayout.java

for sched* events, you will populate all the views.

• How to create a custom parser/analyzer in Trace Compass if needed

You can, we have tutorials here:

https://github.com/dorsal-lab/EclipseTrainingSlides

and here

https://archive.eclipse.org/tracecompass/doc/stable/org.eclipse.tracecompass.doc.dev/Developer-Guide.html

But you may be able to get away with using vanilla trace compass.

Please keep us in the loop! I'd love to see some cool results! 🙂

Matthew

---

From: tracecompass-dev <tracecompass-dev-bounces@xxxxxxxxxxx> on behalf of Ahmed Saeed via tracecompass-dev <tracecompass-dev@xxxxxxxxxxx>
Sent: Monday, October 7, 2024 8:05 AM

To: tracecompass-dev@xxxxxxxxxxx <tracecompass-dev@xxxxxxxxxxx>

Cc: Ahmed Saeed <ahmed.saeed@xxxxxxxxxxxxxxxxxxxxxxxx>
Subject: [tracecompass-dev] [Question] Using Trace Compass with Custom Scheduler on TMS570LS1114 MCU

 

Question: I am working on implementing trace visualization for a custom scheduler running on a TMS570 microcontroller. I need guidance on:

1. What is the recommended approach for:
• Collecting trace data from a custom scheduler on an embedded target
• Converting the collected data into a format that Trace Compass can understand
• Visualizing task switches and execution times
2. Specifically, I need to know:
• What trace data format would be most suitable for this use case
• How to implement the necessary trace points in the scheduler
• How to create a custom parser/analyzer in Trace Compass if needed

Current Setup:

• Custom scheduler running on TMS570 MCU
• safe RTOS tracealyzer 

Additional Questions:

1. Are there any existing examples of using Trace Compass with custom RTOS/schedulers?
2. What are the minimum trace data points needed to create a meaningful visualization?
3. Are there any performance considerations when implementing tracing on an embedded target?

Thank you for your help.

Ahmed Saeed,

Embedded SW Engineer

+2 (0)  1096886761

ahmed.saeed@xxxxxxxxxxxxxxxxxxxxxxxx

Coretech Innovations EGYPT
Smart Village - LINX Business Park
Building F5 - (A05) - Giza - 12577