Manual Pages  — FFCLOCK

The ntpd(8) daemon has been the dominant solution for system clock synchronisation for many years, which has in turn influenced the design of the system clock. The ntpd daemon implements a feedback control algorithm which has been demonstrated to perform poorly in common use cases.

Feed-forward clock synchronisation algorithms implemented by an appropriate daemon, in concert with the FFCLOCK kernel support, have been shown to provide highly robust and accurate clock synchronisation. In addition to time keeping, the FFCLOCK kernel mechanism provides new timestamping capabilities and the ability to use specialised clocks. Feed-forward synchronisation is also very well suited for virtualised environments, reducing the overhead of timekeeping in guests and ensuring continued smooth operation of the system clock during guest live migration.

The FFCLOCK kernel support provides feed-forward timestamping functions within the kernel and system calls to support feed-forward synchronisation daemons (see ffclock(2) ).

When feed-forward clock support is compiled into the kernel, multiple system clocks become available to choose from. System clock configuration is possible via the kern.sysclock sysctl(8) tree which provides the following variables:

kern.sysclock.active
	Name of the current active system clock which is serving time. Set to one of the names in kern.sysclock.available in order to change the default active system clock.
kern.sysclock.available
	Lists the names of available system clocks ( read-only ).

Feed-forward system clock configuration is possible via the kern.sysclock.ffclock sysctl tree which provides the following variables:

kern.sysclock.ffclock.version
	Feed-forward clock kernel version ( read-only ).
kern.sysclock.ffclock.ffcounter_bypass
	Use reliable hardware timecounter as the feed-forward counter. Will eventually be useful for virtualised environment like xen(4), but currently does nothing.

clock_gettime(2), ffclock(2), bpf(4), timecounters(4), sysctl(8)