FW: [e2e] High Packet Loss and TCP

[Jose L Gil Gonzalez]

i believe that the mapping between an average bit error rate (BER) and
an average packet loss rate (PLR) (at the IP level) requires to
specify the bit error distribution. 

In the mobile comms link example there are several bit error
distributions produced by different propagation profiles (Rayleigh,
Rician, etc). A 10-6 BER does not produce the same packet loss rate
for all propagation conditions. Additionally, as David Reed says,
these technologies have FEC and ARQ mechanisms (and other features)
that recover from BER differently (for different BER distributions).
If you have to work with TCP over mobile comm links i would consider
the propagation profile before mapping BER into PLR.

Jose


>-----Original Message-----
>From: Greg Minshall [mailto:minshall at acm.org]
>Sent: 02 May 2003 18:00
>To: Jonathan M. Smith
>Cc: Ross Callon; end2end-interest at postel.org; kurose at cs.umass.edu;
ilija
>Hadzic
>Subject: Re: [e2e] High Packet Loss and TCP 
>
>
>am i right that (10**-6) is order of magnitude 10% packet loss (+/-,
>depending 
>on packet size)?  if so, theory and practice and whatever seem to
converge.
>
>always a nice (if somewhat suspicious) outcome.
>