Open Geal opened 6 years ago
not much we can do from sozu's code, if the kernel suddenly decides to use the HPET clock, everything will get slow. Maybe having a configuration check somewhere at startup?
Maybe having a configuration check somewhere at startup?
Good idea. Could also be part of some kind of a "generate troubleshooting information" tool / command.
In a stress test, I got the following profiling info:
the calls to
sys_clock_gettime
corresponds to request for the monotonic time, to measure timeouts, service time, response time, etc. I was not really expecting this to be a performance issue, so it might be useful to investigate.I saw this in dmesg:
So is the tsc based clock source really unreliable?