On Thu, 4 Oct 2012 13:04:27 -0400 Bob Bownes <bow...@gmail.com> wrote:
> The problem stems from one of the two (identical) machines drifting off by > 60-70 seconds per day. So a few ms here and there are ok. Is it drifting without ntp or with ntp? 1minute drift per day is not unheard of for standard PC RTCs.. i've seen even worse.. > This is all relative to two internal stratum 2 servers on the same command > and control network. No large xfers allowed over it. There are physically > separate data, backup, and application networks for that. On the average LAN (ie one that is connected by a couple of switches. No routers inbetween), the RTT will be dominated by the minimum packet times of ethernet. That's where the 200us delay comes from (actually it's below 1us for Gbit, but for some reason i've never seen it go under 100us). With modern switches that do worm-hole routing (ie just use the destination address in the first few bytes and then pass the packet on without further delay, while the "end" of the packet is still on the wire and not yet received by the switch) you dont even get much added delay from using multiple switches in line. With such large delays you can ignore any wire delay completely (which is in the range of 1us for maximum length cable). With routers it looks a bit different, there the whole packet is first stored in memory before being processed (due to more complicated routing decision). Due to this, RTT is still dominated by delay of the network hardware and not speed of light. Actually speed of light related delays will be buried deep in the noise unless you go trans-continental. And even then, router delay will still dominate. (eg RTT Europe-Japan is around 500ms, by which time a packet would be half way to the moon) So, if you have any significant time difference (>1ms) between two systems that synchronize to the same NTP server in the same LAN, then the problem lies somewhere else than the network. Attila Kinali -- There is no secret ingredient -- Po, Kung Fu Panda _______________________________________________ time-nuts mailing list -- time-nuts@febo.com To unsubscribe, go to https://www.febo.com/cgi-bin/mailman/listinfo/time-nuts and follow the instructions there.