[time-nuts] Designing an embedded precision GPS time server
denny at cococafe.com
Fri Oct 27 18:09:57 EDT 2017
> On Oct 26, 2017, at 19:29, Chris Caudle <chris at chriscaudle.org> wrote:
> On Thu, October 26, 2017 7:38 pm, Denny Page wrote:
>> If you are going to do PTP with ptp4l, or NTP with Chrony, you are going
>> to want hardware timestamping support on the ethernet phy.
> Or the MAC. The processor used on BeagleBone Black has timestamping in
> the MAC. Not quite as accurate as stamping in the phy, but should be a
> relatively consistent fixed offset.
Yes, but it might not be as consistent as you would like. The Intel i210 is a pretty good reference. It does mac timestamping and has a built-in phy. At 100Mb, Intel advertises a timestamp latency range of 984-1024ns for outbound packets, and 2148-2228ns for inbound packets. The ranges are a result of mac clock issues, unassociated with phy communication.
External measurement shows that the mean values are actually outside the ranges given, 1044ns for outbound, and 2133ns for inbound. [I suspect, but don’t know for sure, that the offset is a result of a fixed 20ns delay between the phy and mac.] Anyway, on a loopback between two i210 instances, you will see an average total timestamp latency of 3177ns, with a standard deviation of around 100ns. Pretty good, but not that great.
What chip does the BeagleBone Black use? Do they publish specs on the ethernet timestamp latency?
More information about the time-nuts