Feedback on draft-ietf-dccp-tfrc-voip-05.txt

A few days ago, I sent you email saying the following:


Ladan:
> The draft is not clear on the when feedback is sent
back in instances 
> where the loss intervals are at most 2RTTs  and the
loss event rate is 
> computed as K/N? For TFRC, once a packet is lost in a
new RTT, the 
> loss event rate changes, feedback is sent immediately
and the 
> remaining losses within that RTT are ignored. However
for TFRC-SP, 
> each loss modifies the loss event rate, therefore,
should feedback be 
> sent immediately and on every loss (since p changes
with every loss) 
> or should it wait until the end of the RTT?

Sally:
Good question!  Feedback should be sent when the loss is
first 
detected, and every
succeeding RTT.  I added a section to the draft discussing
this, and it 
is appended to
the end of this email.  Feedback would be welcome.


Since then, I have made a change to TFRC-SP to deal with
possible 
changes in the
loss event rate with short intervals.  So the text attached
to the 
earlier email has
been replaced by the following.

Many thanks for the feedback.

- Sally
http://www.icir.org/floyd/


Added to Section 3:

        Section 5.4 of RFC 3448 specifies that the
calculation of the
        average loss interval includes the most recent loss
interval only
        if this increases the calculated average loss
interval.  However,
        for TFRC-SP, including the most recent loss interval
can increase
        the calculated average loss interval too much if the
most recent
        loss interval ends up being a short loss interval
with multiple
        packet drops.  Therefore, TFRC-SP adds the
restriction that the
        calculation of the average loss interval can include
the most
        recent loss interval only if more than two
round-trip times have
        passed since the beginning of that loss interval.

...
4.6.  The Calculated Loss Interval Length for Short Loss
Intervals

     For a TFRC-SP receiver, the guidelines from Section 6
of RFC 3448
     govern when the receiver should send feedback messages.
 In
     particular, from [RFC3448], "a feedback packet
should ... be sent
     whenever a new loss event is detected without waiting
for the end of
     an RTT".  In addition, feedback packets are sent
at least once per
     RTT.

     We note that in a connection with a short loss interval
(less that
     two round-trip times) and multiple packet losses per
loss interval,
     the loss interval length calculated for the most recent
loss
     interval can decrease from one round-trip time to the
next, as
     multiple packet losses are detected for that loss
interval.  As an
     example, when the TFRC-SP receiver sends a feedback
packet the
     current loss interval might contain N packets, with
only one loss,
     giving a calculated loss interval length for that
interval of N
     packets.  When the TFRC-SP receiver sends a feedback
packet one
     round-trip time later, K additional lost or marked
packets might
     have been detected, giving a calculated loss interval
length for
     that interval of only (N+K)/(K+1) packets.  For K=N/2,
this could
     lead to a change in the calculated loss interval length
from N to
     close to 2 packets.  To prevent unnecessary
oscillations in the
     average loss interval, Section 3 specifies that the
current loss
     interval can be included in the calculation of the
average loss
     interval only if the current loss interval is longer
than two round-
     trip times.  For a loss interval longer than two
round-trip times,
     the detection of new losses for the loss interval will
not
     *decrease* the calculated loss interval length for that
loss
     interval.

Sally:

This definitely clarifies when feedback is sent by the
receiver. But  
I think the whole discussion on the short intervals and when
the are  
used in the average loss interval is fairly complicated. 
I've read  
it a few times and I'm still not sure I get it  It would
be great  
if you could add pseudo-code similar to that of RFC 3448 to
this  
draft, as it would eliminate all possible ambiguities on how
I_tot is  
computed.

Ladan


On Oct 23, 2006, at 1:15 AM, Sally Floyd wrote:

> A few days ago, I sent you email saying the following:
>
>
> Ladan:
>
>> The draft is not clear on the when feedback is sent
back in  
>> instances where the loss intervals are at most
2RTTs  and the loss  
>> event rate is computed as K/N? For TFRC, once a
packet is lost in  
>> a new RTT, the loss event rate changes, feedback is
sent  
>> immediately and the remaining losses within that
RTT are ignored.  
>> However for TFRC-SP, each loss modifies the loss
event rate,  
>> therefore, should feedback be sent immediately and
on every loss  
>> (since p changes with every loss) or should it wait
until the end  
>> of the RTT?
>>
>
> Sally:
> Good question!  Feedback should be sent when the loss
is first  
> detected, and every
> succeeding RTT.  I added a section to the draft
discussing this,  
> and it is appended to
> the end of this email.  Feedback would be welcome.
>
>
> Since then, I have made a change to TFRC-SP to deal
with possible  
> changes in the
> loss event rate with short intervals.  So the text
attached to the  
> earlier email has
> been replaced by the following.
>
> Many thanks for the feedback.
>
> - Sally
> http://www.icir.org/floyd/

>
> Added to Section 3:
>
>        Section 5.4 of RFC 3448 specifies that the
calculation of the
>        average loss interval includes the most recent
loss interval  
> only
>        if this increases the calculated average loss
interval.   
> However,
>        for TFRC-SP, including the most recent loss
interval can  
> increase
>        the calculated average loss interval too much if
the most  
> recent
>        loss interval ends up being a short loss
interval with multiple
>        packet drops.  Therefore, TFRC-SP adds the
restriction that the
>        calculation of the average loss interval can
include the most
>        recent loss interval only if more than two
round-trip times  
> have
>        passed since the beginning of that loss
interval.
>
> ...
> 4.6.  The Calculated Loss Interval Length for Short
Loss Intervals
>
>     For a TFRC-SP receiver, the guidelines from Section
6 of RFC 3448
>     govern when the receiver should send feedback
messages.  In
>     particular, from [RFC3448], "a feedback packet
should ... be sent
>     whenever a new loss event is detected without
waiting for the  
> end of
>     an RTT".  In addition, feedback packets are
sent at least once per
>     RTT.
>
>     We note that in a connection with a short loss
interval (less that
>     two round-trip times) and multiple packet losses
per loss  
> interval,
>     the loss interval length calculated for the most
recent loss
>     interval can decrease from one round-trip time to
the next, as
>     multiple packet losses are detected for that loss
interval.  As an
>     example, when the TFRC-SP receiver sends a feedback
packet the
>     current loss interval might contain N packets, with
only one loss,
>     giving a calculated loss interval length for that
interval of N
>     packets.  When the TFRC-SP receiver sends a
feedback packet one
>     round-trip time later, K additional lost or marked
packets might
>     have been detected, giving a calculated loss
interval length for
>     that interval of only (N+K)/(K+1) packets.  For
K=N/2, this could
>     lead to a change in the calculated loss interval
length from N to
>     close to 2 packets.  To prevent unnecessary
oscillations in the
>     average loss interval, Section 3 specifies that the
current loss
>     interval can be included in the calculation of the
average loss
>     interval only if the current loss interval is
longer than two  
> round-
>     trip times.  For a loss interval longer than two
round-trip times,
>     the detection of new losses for the loss interval
will not
>     *decrease* the calculated loss interval length for
that loss
>     interval.
>
>

Ladan -

> This definitely clarifies when feedback is sent by the
receiver. But I  
> think the whole discussion on the short intervals and
when the are  
> used in the average loss interval is fairly
complicated.  I've read it  
> a few times and I'm still not sure I get it  It would
be great if  
> you could add pseudo-code similar to that of RFC 3448
to this draft,  
> as it would eliminate all possible ambiguities on how
I_tot is  
> computed.

Good idea.  I will add the pseudocode.  It is a small change
to this in  
Section 5.4
of RFC 3448:

------------------------------------------------------------
------------ 
-------------------------------
    Thus if the most recent loss intervals are I_0 to I_n,
with I_0 being
    the interval since the most recent loss event, then we
calculate the
    average loss interval I_mean as:

       I_tot0 = 0;
       I_tot1 = 0;
       W_tot = 0;
       for (i = 0 to n-1) {
         I_tot0 = I_tot0 + (I_i * w_i);
         W_tot = W_tot + w_i;
       }
       for (i = 1 to n) {
         I_tot1 = I_tot1 + (I_i * w_(i-1));
       }
       I_tot = max(I_tot0, I_tot1);
       I_mean = I_tot/W_tot;

    The loss event rate, p is simply:

       p = 1 / I_mean;
------------------------------------------------------------
------------ 
----------------------------------

The pseudocode for tfrc-sp would just say:
   If the current loss interval I_0 is "short"
     then I_tot = I_tot1;
     else I_tot = max(I_tot0, I_tot1);

- Sally




- Sally
http://www.icir.org/floyd/