On 25/01/13 16:13, Andreas Schultz wrote:
Only with the patch applied, I didn't investigate it further but my guess is that with real-time scheduling the events can happen out-of-order at some point.
Yes. Strange.
Btw, the assertion in question just checks whether time measurements during latency analysis happen in correct order (nn_send() goes first, etc.)I did remove the assertion, but the results where completely off then.
I've applied the patch. While the test failed in some 50% of runs, it still did deliver results in the remaining cases. The latency seems to look worse rather than better. See the attached graph.
Martin
Attachment:
inproc_lat.png
Description: PNG image