Re: [AVT] How is SDP a=fmtp used for RFC 2198?

"Gunnar Hellstrom" <gunnar.hellstrom@omnitor.se> writes:
>3. Each generation of redundant info gives quite good reliability increase.
>
>Look at this calculation:
>
>For RFC2793bis text transport in text form, this is the theory:
>
>Assume that the risk of loosing a packet is p
>The risk of loosing also next text packet is also p
>
>The bad case is only when both are lost. The likelihood for that is p*p .
>
>You can check with 50% loss and sending two packets. There are totally four
>outcomes.
>
>lost -not lost
>lost - lost
>not lost - lost
>not lost - not lost
>
>Only one out of four is unsuccessful. that is 25%.
>
>0.5 * 0.5 = 0.25 the probability calculation is valid.
>
>If you add one more redundancy level, the likelihood of all three lost is
>p*p*p
>
>With real figures, 20% packet loss and three transmissions give 0.2 * 0.2 *
>0.2 = 0.008 =0.8%total loss.

        This assumes totally independant loss.  The reality is that loss is
pretty strongly _not_ independant; i.e. the conditional probability of
losing packet N+1 after losing packet N is high, and typically way higher
than the base loss probability p (probability of losing packet N if packet
N-1 was received).  I.e. bursts are not an unusual type of loss.

        From numbers I've seen, the conditional probability for a second
packet to be lost (loss burst size >= 2) on the internet is around 15% or
more (with a p~= 1%).

        Of course these values are way lower than the 50% or 20% mentioned
above, but they're also heavily dependant on the physical links traversed.
(a 3GPP link may be much worse).

-- 
Randell Jesup, Worldgate Communications, ex-Scala, ex-Amiga OS team
rjesup@wgate.com
"The fetters imposed on liberty at home have ever been forged out of the weapons
provided for defence against real, pretended, or imaginary dangers from abroad."
		- James Madison, 4th US president (1751-1836)


_______________________________________________
Audio/Video Transport Working Group
avt@ietf.org
https://www1.ietf.org/mailman/listinfo/avt