I *think* you are suggesting, the following, but would like to be sure:
* Feedback SHOULD/MUST NOT be less than once per RTT
* Feedback MAY be more,
but SHOULD be less than X percent of forward bandwidth?
Have I understood?
We are refining TFRC at this moment. Although may not be able to
pin-down the spec. based on currenty experience, IMHO it would be great
if the wording of TFRC.bis (3448.bis) were to be compatible with these
ideas.
Gorry
Ingemar Johansson S (LU/EAB) wrote:
Hi
One issue that may be important to know in this aspect is that there
will probably need to be some limit to how often feedback is transmitted
esp. for VoIP applications with low birate.
Low bitrate VoIP typically have a bitrate of 5-12kbps (payload bitrate)
and in 3GPP the bearers that are specified (today) for these services
have an RTT in the range 200-700ms depending on load and coverage
situation. One of the worries regarding DCCP is that feedback cost too
much and having more than one feedback per RTT will further increase the
system load.
One solution may be that an extra condition is added that specifies that
feedback should not consume more than X percent of the total transmitted
bitrate combined with the rule that feedback should be transmitted
atleast once per RTT.
Maybe a little odd formulation but hopefully the intention is
understood. Anyway, with this rule feedback would not consume an
unreasonable amount of BW for low bitrate VoIP applications.
Ingemar
-----Original Message-----
From: Sally Floyd [mailto:sallyfloyd@xxxxxxx]
Sent: den 20 augusti 2007 02:27
To: Eddie Kohler
Cc: Gorry Fairhurst; Golam Sarwar; Sebastien Ardon; 'dccp'
working group; Roksana Boreli
Subject: Re: DCCP for VoIP
This sounds like a very interesting idea. While the
documents specify
that feedback is sent once per RTT, since this is what TFRC
recommends, I think more frequent feedback is a fine idea even for
low-loss-rate/low-RTT scenarios (twice or four times per
RTT, say?),
and probably mandatory for high-RTT scenarios. My guess is Sally
would agree. Perhaps someone could suggest a change to the CCID3/4
language.
TFRC (RFC 3448) only has a smalll discussion about sending
multiple feedback packets per RTT. E.g., in Section 6:
"If the sender is transmitting at a high rate (many
packets per RTT)
there may be some advantages to sending periodic
feedback messages
more than once per RTT as this allows faster response to
changing RTT
measurements, and more resilience to feedback packet loss."
TFRCbis (draft-ietf-dccp-rfc3448bis-02.txt) adds more
clarification about what needs to be done when the receiver
sends multiple feedback
packets per RTT. From feedback from Arjuna. In Section 6:
"If
the receiver was sending k feedback packets per RTT, step (4) of
Section 6.2 would be modified to set the feedback timer to expire
after R_m/k seconds. However, each feedback packet would still
report the receive rate over the last RTT, not over a fraction of
an RTT."
I think it is good to allow multiple feedback packets per RTT.
However, I personally wouldn't be inclined to make it the
default recommendation for all scenarios, for CCID 3/4, until
there has been a little more experience with it. Just in
case there is some drawback or implementation issue that pops up...
- Sally
http://www.icir.org/floyd/
