[ntp:questions] On flakeways, scattergrams and timequakes
David L. Mills
mills at udel.edu
Thu Jan 24 19:59:43 UTC 2008
Your comment on another thread about weather modelling stirred my pot. I
gave a long thought about modelling when designing the synthetic sources
for the ntpd simulator. With the Allan characteristic in mind the
synthetic sources use exponentially distributed phase noise plus
random-walk frequency noise, which nicely models the Allan
characteristic shown in my papers and others.
But, that's kid's stuff. The real ornery cases can't be explained this
easily. The best examples I have were when ARPAnet was operating really
badly, but nobody knew why. It turned out routes were flapping between
landline and satellite paths, as confirmed by the NTP rawstats wedge
scattergrams. The scattergrams showed four overlapping wedges, each
corresponding to one possible combination of landline and satellite
links on each direction.
Another interesting case was the early Internet path via the Atlanic to
Italy, which had frequent glorious flaps over one second. For that
matter the huff-n'-puff data shown in das Buch are wonderful grist for
the NTP mill.
Once upon a time earlly in Internet times Bob Braden at UCLA operated a
"flakeway" best described as a route deflector that scattered incoming
packets to random destinations. It was used as for testing TCP and
routing algorithms. We need a similar package that can operate as an NTP
server and do various kinds of timequakes for testing.
More information about the questions