Re: [dccp] Re: [AVT] Would DCCP kill Internet real-time media distribution?

[Jon Crowcroft <Jon.Crowcroft@cl.cam.ac.uk>]

In missive <web-2064542@multicasttech.com>, "Marshall Eubanks" typed:

 >>On the Internet2, I would estimate that a substantial fraction of the loss that I
 >>see in multicast UDP streaming is not due to congestion, in many cases > 50%. How
 >>would DCCP respond to such false congestion signals ? Is there a CETEN like mechanism
 >>to deal with this ?

1/ do you mean half the time that you see loss, or that you see 50% loss? i'venever seen any loss level like that
except on a severaly broken network, where no amount of dumb CBR streamed UDP or  clever adaptive dccp will be any
use nor will TCP btw

2/ deploy ECN. that will let you cope with non congestive loss through other means - hey, >10% loss is the
Internet's way of just telling you "NO"

[its called CALL BLOCKING in phone nets, but people just dont seem to get that:-)

btw the comment on adaptive rate sucks - its correct - the usertrials folks at UCL did certainly show that there is
little point increasing the quality of video/audio if you are going to have to decrease it , within a session, as
the user experiecne contains a lot of hysteresis - i.e. they habituate, but the user reports the worst rate they
see, and variation (up from that and down to it) degrade the overall experience more (even though the average rate
may be better)....

i think this points the way (sorry to say it) to a model where we do admission control for streaming based on large
aggregate measurements and ECN, and not to rate adaption but i know this is heretical

technically, one can estimate a rate over the multiple RTT time scale via TFRC but the question is what do do about
large scale fluctuation in the load at a bottlenect ...and it seems to me that we dont have a story here yet, even
though we've been muckng about with this stuff for about 15 years...

cheers

j

_______________________________________________
dccp IETF mailing list: dccp@ietf.org
list info:  https://www1.ietf.org/mailman/listinfo/dccp
wg charter: http://www.ietf.org/html.charters/dccp-charter.html