Closed mansr closed 5 years ago
This appears to be caused by the new kernel config having HZ=1000. Lowering HZ to 100 makes the problem go away.
for the 1000Hz issue; what I understand is the following:
Mans, will do any other work related to this, which would be testing the wifi and also testing the canbus at full load. Anything else needed?
And is setting to 100 and then testing it indeed the way to go? What are the other machines at?
I don’t know anything about all this. Just want to make sure its properly solved and preferably with as little time from Jeroen as possible
Mans wrote:
the other machines use dynticks/nohz they are also much faster
Jeroen wrote:
the other machines are not so relevant, since this patch tries to workaround a bug somewhere else in the ccgx.. I am not aware other kernels have this issue. (it might be fixed by the mutex issue as well ...) test it with cansequence.c the CAN-bus is the most time critical function I am aware we have... And it is designed around the assumption that there are no messages dropped
For Groningen Hub1 this seems to help, lets wait for https://groups.google.com/d/msg/victron-dev-venus/RluRx3xcl5Q/XMmxAFzJAQAJ or otherwise close the issue. No hung processes spotted yet.
User reports:
I updated to 2.30~35 two days ago and have had no reboot since. I can also confirm that the average load is back to normal levels on my system.
Indeed; closing!
Report of increased CPU usage in 2.30:
v2.30\~27:
v2.20~67:
The main difference here seems to be the GUI, going from 11% CPU to 26%.