On Feb 10, 2008 5:12 PM, Dag-Erling Smørgrav <[EMAIL PROTECTED]> wrote: > Assuming the measuring server correctly estimates and compensates for > network latency (using the very simple algorithm described in the NTP > and SNTP RFCs), it should not affect the offset it sees unless there is > considerable asymmetry between the outbound and inbound leg. >
There is, in my experience, almost always asymmetry on any internet route of appreciable length. ISPs generally use "hot-potato" routing, meaning they dump packets bound for another ISP at the closest peering point. This is basically true all the way up to and including the Tier-1 ISP level. So unleess the closest peering point between two networks is the same in both traffic directions (e.g. in the same major city, but on different ISP networks), you will see route asymmetry. You can test this yourself by using any of the freely available web-based traceroute tools. Traceroute TO your server FROM the tool, and then do the reverse FROM your server TO the website. The two routes are usually quite different. This is one design assumption made in NTP that seems to be basically wrong in the general internet case, but I don't have any suggestions about how to get around it with the current ntp wire protocol. -- RPM _______________________________________________ timekeepers mailing list [email protected] https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers
