Giacomo,
IMHO, it depends...on MTU, cell size (ATM) and type of FEC.
FEC type: if you use Viterbi only, the errors come in bursts of about 20
bits, so, for BER=10E-6 you'll have one loss of 20 bits every 20*10E6.
That could affect a maximum of 2 ATM cells and there are 23,585 cells in
20*10E6 bits. Again, those 2 ATM cells can affect up to two TCP
segments, let's say of 1,500 Bytes=12,000 bits. In 20*10E6 bits there
are 1,667 segments then, and you're going to re-transmit every 1,667, up
to two segments plus, all those already arrived but not acknowledged,
depending on CWND, maybe another 4. As we can see, at BER>10E-4 the
system just freezes.
If transmission is Frame Relay, you lose who knows what amount of
segments.
Now, if FEC's are concatenated, with Reed Solomon, the errors are more
evenly distributed and you may expect losing, with BER=10E-6, one
segment every 10E6 bits that contain 83 segments of 1500 Bytes. But you
don't see Reed Solomon implemented networks at less than BER=10E-8 and
at that rate, the loss would be one every 8,300.
Best regards,
Mike
Miguel A. Checa
ppal. engr.
Gulf Communications International, Inc.
--------------------------------------
Giacomo Morabito wrote:
>
> Please, I have a question.
>
> What are the typical values of the "packet loss probabilities"
> due to link errors in satellite links, or, alternatively,
> what are the values of the packet loss probability corresponding
> to a BER going from 10^{-9} to 10^{-4}?
>
> I hope that someone can help me.
>
> Best regards,
> Giacomo
>
> -----------------------
> Dr. Giacomo Morabito
> Research Engineer
> Broadband & Wireless Networking Laboratory
> School of Electrical and Computer Engineering
> Georgia Institute of Technology
> Atlanta, GA 30332
>
> Tel.: (404) 894-6616
> Fax.: (404) 894-7883
> E_mail: [email protected]
> http://www.ee.gatech.edu/~giacomo
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:58 EST