AM6441: CODESYS EtherCAT Master Max jitter

Ivan Stoyanov

Part Number: AM6441
Other Parts Discussed in Thread: SK-AM64B, TMDS64EVM

Tool/software:

Hello,

We are running some performance measurements on AM6441 processor. For that purpose we are using latest Arago for building base Linux image with RT patch applied. The results from cyclictest are:

cyclictest -l100000000 -m -p99 -c0 -t -i200 -h100

# Total: 100000000 099999997
# Min Latencies: 00006 00006
# Avg Latencies: 00008 00008
# Max Latencies: 00122 00165
# Histogram Overflows: 00000 00000
# Histogram Overflow at cycle number:
# Thread 0:
# Thread 1:

We have ran tests with CODESYS EtherCAT Master to measure max cycle time of EtherCAT task but the results are also interesting. We see jumps of the max cycle time as follow.

We have followed the best practices from the document below but without obvious improvement of the jitter and EtherCAT Master max cycle time.

Performance Metrics of AM6x Processors as CODESYS EtherCAT Controller

We have ran the same tests on SK-AM64B and TMDS64EVM.

Does anyone observe such Max jitters and are they normal for this processor?

Does anyone observe same behavior? Any ideas how to improve it?

Thank you in advance!

Best regards,

Ivan

8 months ago

0 Daolin Qiu 8 months ago

TI__Genius 17165 points

Hello Ivan,

Ivan Stoyanov said:
For that purpose we are using latest Arago for building base Linux image with RT patch applied. The results from cyclictest are:

cyclictest -l100000000 -m -p99 -c0 -t -i200 -h100

When you say "latest Arago for building base Linux image" is this the latest image to the TI Processor-SDK RT-Linux? Which Kernel version is this?

Are the results from cyclictest after running stress-ng in the background (as specified in the Performance Metrics of AM6x Processors as CODESYS EtherCAT Controller app note)?

It looks like according the latest SDK 10.1, the cyclictest results after 6 hours is shown in https://software-dl.ti.com/processor-sdk-linux/esd/AM64X/latest/exports/docs/devices/AM64X/linux/RT_Linux_Performance_Guide.html. On SDK 9.2.1.10, the max latency result looks to be better: https://software-dl.ti.com/processor-sdk-linux/esd/AM64X/09_02_01_10/exports/docs/devices/AM64X/linux/RT_Linux_Performance_Guide.html?highlight=cyclictest

Ivan Stoyanov said:
We see jumps of the max cycle time as follow.

Are you able to view the cycle time in real-time like the graph shown in the app note? The steps to set this up is summarized in the 7.5 How to View Performance Measurements section, specifically "To graphically view cycle time and jitter statistics in real-time". This would give a better indication to see if the cycle time is jumping periodically like what was observed in the app note.

Do you have a codemeter license application running on the AM64x device you are testing? I'm guessing if you had run this for longer than 30 min, you would have needed a license to enable that. We noticed that the license application impacts the cycle time periodically, explained in the app note.

How long have you run CODESYS for these results? Have you monitored the CPU load during the same time?

-Daolin

Processors

Processors forum

AM6441: CODESYS EtherCAT Master Max jitter