My PTP master is now (sort of) working and synchronizing a slave. I don't have any PTP capable switches (anybody want to send me one?), so the master to slave latency is noticeable. Initially I had a Netgear gigabit switch (GS605) and a linksys wireless router in the path between the two boards. The latency was (apparently) around 19µs. This seemed high to me as I was expecting around 5µs for the path to the switch (minimum packet size at 100Mbps) and then maybe 5µs for the connection between the switches and the rest has to be in the switch actually forwarding the packet. The boards themselves time the start of the packet for send and receive. Anyway, I cut the linksys router out of the path, and the latency dropped to 10µs. This implies a switching latency of 5µs plus a transmission time of 5µs. The documentation for the switch is pretty useless, so it isn't clear if this is right.
The servo loop in PPSI seems to oscillate (by comparison with the NTP PPS loop) and after 30 minutes it has not yet settled down. It varies by at least a microsecond from the real time (as determined by another GPS PPS pulse). I think that the servo code needs some work.
**UPDATED**
After almost 24 hours, the offset is still varying by a significant amount -- up to a microsecond. It isn't clear whether this is actually reflective of some underlying variation in the system or just the servo loop.
No comments:
Post a Comment