[e2e] Re: crippled Internet
J.Crowcroft at cs.ucl.ac.uk
Thu Apr 26 00:16:14 PDT 2001
In message <200104260527.f3Q5R9G13202 at calcite.rhyolite.com>, Vernon Schryver ty
>>> From: Randy Bush <randy at psg.com>
>>> so the jitter characterization, "50ms or so of frequent variation," would
>>> seem to be the interesting issue. what is "frequent variation," and how
>>> should it be measured? i.e. if 3% of the samples are outliers, how they
>>> are distributed in time would seem to be critical.
>>Given the sort of digital processing against echos that is now routinely
>>done for good speakerphones and the CPU cycles available in modern personal
>>computers, I wonder if the old worries about echo cancelers are irrelevant.
>>There's also the fact that everything is digital in VoIP.
experience with actually using audio tools is that generally the net
is not the main problem so long as neither end is dial up and both
ends haev good analog audio setups - can cite lots of papers pointing
at the problems mostly being with mikes and speakers - typical tier 1
and 2 paths are fine when there isnt inter-provider congestion...
(i.e tier 0:-) and delay variance for the 9th centile is usally fairly
bounded (the point of doign loss concealment in rat for example,
is to deal with
i) the occasional congestion burst and noise loss
ii) the 5-10% of packets in the tail of the delay distributuion caused
by the mix of tcp and udp arrival processes in the queues -the number
of flows is getting largew enough in a lot of paths with enough
variance o nthe rtts and tail circuits that this distribution is
tending to what you would expect with any large scale random stats so
this is starting to work pretty well more and more of the time in my
again a lot of blame on poor experience can be pointed at the audio
driver, OS delays and poor analog and a2d setup in end systems (and
serialisatioin/modem delay in dial up which makes life for voip a complete
i actually think christian is right that we could deploy jean-yves'
idea (or something like it) actually JUST in the inter-domain points
and access ISPs would be enough nowadays imho - but of course i can't
prove this would work (at least not for randy:-)
More information about the end2end-interest