[Interest] Packet arrival-time resolution? QUdpSocket
Thiago Macieira
thiago.macieira at intel.com
Thu Nov 30 17:27:36 CET 2017
On quinta-feira, 30 de novembro de 2017 07:49:15 PST Jason H wrote:
> The speed of light is ~1ft per ns, and your clock is running at ~0.33-1 ns.
Uh... that's not the same clock.
You're mistaking the CPU cycle clock with the timestamp counter and the real-
time clock. They're not the same. Neither runs at 1 ns resolution (the RTC
runs MUCH slower). The TSC on modern Intel platforms runs at a constant rate
and is the same in all cores and processors, but that's not a guarantee on
other platforms.
> There's plenty of factors to prevent absolute best performance, not even
> including that the speed of light in a wire is about 1ft per 3ns, due to
> inductance (multi-layer PCBs also slow it down). Meanwhile the speed of
> sound is ~1.1ft per ms at STP.
In more usual units:
speed of light = 299 792 458 m/s = 299.792458 mm/ns
speed of sound at STP = ~340 m/s = ~340 mm/ms
> If I can get reliable 1 ms packet resolution
> and accuracy I'll be fine. I can alter the data and node configuration to
> support some jitter. As long as the hosts accurately record the event on
> local clock time, and I know the offsets of each clock, can calculate the
> true event time. If they are all sub ms deviation, then I can take it as-is
> and know I'll be within 13 inches.
Unless the wall-clock suffers a time jump. That's where I thought the
monotonic time would have been useful. You could convert it to wall-clock at
your leisure.
--
Thiago Macieira - thiago.macieira (AT) intel.com
Software Architect - Intel Open Source Technology Center
More information about the Interest
mailing list