[ntp:hackers] Asymmetric delays; leapsecond stuff
David L. Mills
mills at udel.edu
Sun Jul 8 03:05:38 UTC 2007
Guys,
1. Changing the leapseconds command code to something else proved a
disaster. The original version would either hang up in a request/reply
loop or reset due invalid command code. So, I changed it back to the
original. While the original code will not import the TAI from a server,
it will set it correctly if the leapseconds file is present.
2. There is some experimental code in ntp-dev designed to deal with
asymmetric delays. It does this by calculating the rates for both the
outbound and inbound data stream. To do this it keeps track of the
differences between timestamps for different packet lengths. At the
moment this requires Autokey, as this is the only function that uses
packets of different lengths.
So, if you bring up Autokey you should see two new additions to the ntpq
billboards, in=number and out=number. The numbers are in kB/s. At the
moment nothing is done with these numbers; further plans are to groom
for noise, remove spikes, make it work with other than Autokey and then
use the result to correct significant asymmetric delays.
Dave
More information about the hackers
mailing list