1. clock - Wall clock; Used to represent date and time; After synchronization, the local clock of our server can jump backward or forward in time - Monotonic clock; Used to calculate duration; The time ALWAYS moves forward and will not be impacted by variations leading to jumps in time 2. Wall clock - Linux uses UTC as wall clock - UTC is not accurate because earth's rotation speed varies - UTC does not contain Leap Second, so Linux adjust the UTC time to real earth rotation by inserting leap seconds into UTC time scale 3. Wall clock Synchronization - Normal solution: synchronize with NTP (Network Time Protocol). Because of the network delay, there will be at least a 35 millisecond deviation, and the maximum may exceed 1 second. - Good solution: receive accurate wall time with GPS, synchronize with other servers via PTP, and can achieve sub-microsecond precision - Google solution: Google uses a GPS receiver to receive a standard wall clock, and then deploys the atomic clock through the computer room, so that its accuracy can reach an error of only 1 second every 20 million years, in this way to prevent the failure of the GPS receiver. Send accurate time to other servers with TrueTime API - Google Spanner (distributed database) is based on this. 4. Resolve Leap Second Issues - Bad Solution: notify the kernel, and allow it to insert the leap second, which cause stepping the system clock backwards - Good Solution: Slew the leap second (since RHEL 6.8 and RHEL 7.2), allowing this second to be spread out over a period of time instead of an immediate step back in time 5. How computer manage time? - Clock interrupts is based quartz crystal oscillator. Computer update the wall time during each clock interrupt - How to deal with 1 microseconds (small precision)? BogoMIPS Ref: https://access.redhat.com/articles/15145 https://itnext.io/as-a-software-developer-why-should-you-care-about-the-monotonic-clock-7d9c8533595c https://donggeitnote.com/2021/06/23/clock/