[e2e] How to estimate packet loss?

[Mark Handley]

>In TFRC it is suggested that loss event fraction models TCP more
>accurately than loss fraction does. However I am quite confuse of how to
>estimate packets in a loss interval. Given this data packets, would someone
>please tell me how to estimate packets in a loss interval.
>
>packets sent: 1   2   3   4(loss) 5(loss) 6   7   8   9   10   11   12
>13(loss) 14(loss) 15(loss) 16   17   18   19   20   21(loss)   22   23   24
>
>interval1 = 13 - 6  or interval1 = 13 - 4  ?
>interval2 = 21 - 16 or interval2 = 21 - 13  ?


You don't say anything about the packets/RTT, which determines which
losses are in the same loss event.  So less make some assumptions:

We're transmitting less than 8 and more that 3 packets per RTT, so
there are three loss events in your trace:

  packets 4 and 5 are part of loss event 1
  packets 13, 14 and 15 and part of loss event 2
  packet 21 is part of loss event 3.

Now it doesn't matter all that much exactly how you measure loss, so
long as each packet (lost or otherwise) is part of exactly one loss
interval.

So: interval 1 could include packets 4 to 12 (9 packets)
    interval 2 could include packets 13 to 20 (8 packets)

Hope this helps,

Mark