Torizon real time

Hello,

one of the TorizonCore advantages is that is optimed for realtime. But in the download section of the images reads: “The images with the PREEMPT_RT patch haven’t been fully validated by Toradex. We have only run smoke tests and cyclictest to make sure that the system boots properly and the PREEMPT_RT patches are working on idle conditions.”. For me this is a contradiction.

Are there any plans for the future to better verify real time capability under stress (but without graphics, like the rt-validation containers) ? Another option would be the osadl realtime farm.

As a customer it would be great if you could guarantee a maximum latency (e.g. 100us) with the TorizonCore base image.

Thanks,
Fabian

Greetings @wheeler,

I agree that statement is a bit confusing but let me try to clarify. Currently we have no plans to further develop the real time variant of TorizonCore, though of course this plan could change.

So with our major releases of TorizonCore we do run basic tests like the rt-validation container to verify basic functionality. In general with these tests we’ve seen an average latency of less than 100us and a max latency of less than 500us. These numbers differ though depending on the exact module, I could give you some ballpark numbers if you could share what module of ours you’re working with.

In general though we always suggest to test your particular real-time workload yourself to get a better idea of latency one might expect.

Best Regards,
Jeremias

Hi @jeremias.tx,

thanks for the clarification.

I’m interested in the results for the Verdin iMX8M Plus QuadLite 1GB, if you already did some measurements. I ran the rt-validiation containers on the Verdin iMX8M Plus Quad 2GB.
2849-32704556-64c4-4d84-ad63-2f96f14f363a.png

Is it to be expected that the results are similar among the iMX8M Plus module variants?

Best Regards,
Fabian

Your results seem in-line with our own internal testing.

As for any difference amongst variants. Since the same SoC/CPU is being used there shouldn’t be any difference compute wise. The only exception would be workloads where the difference in memory would be a factor. These cases obviously would favor the 2GB variant, otherwise I wouldn’t expect any noticeable differences.

Best Regards,
Jeremias