Jamshid-
Thank you for raising an important point. Not all bit errors are equivalent.
Most (all?) satellite links utilize FEC. Block coding results in lost blocks
when uncorrectable errors occur. The packet loss rate due to corrupted blocks is
much different that that due to a uniform distribution of bit errors. So, the
translation of channel BER (after FEC) to packet loss rate (which is necessary
to interpret the results of the reference below will depend heavily on the
adaptation of IP to link layer.
Having said all that, my understanding is that to get 'decent' performance out
of TCP (ie., a window > 1) packet losses should be limited to less than 1 per
RTT.
There was a thread recently on the pilc list (see http://pilc.grc.nasa.gov/pilc)
on channel BER that might be of interest to tcpsat participants. I am
crossposting to the pilc list and would like to recommend continuing this
discussion there.
--aaron
From: [email protected] AT mime on 03/31/99 10:23 AM
To: [email protected] AT mime@PMDF-IMP@HUB-H4@CCMAIL
cc: [email protected] AT mime@PMDF-IMP@HUB-H4@CCMAIL (bcc: Aaron D
Falk/ES1/ES/HSC)
Subject: Re: BER and TCP/IP performance
The currently accepted best formula for TCP performance as a function of packet
loss is given in the following paper:
Padhye, J., Firoiu, V., Towsley, D., and Kurose, J., Modeling TCP Throughput: a
Simple Model and its Empirical Validation, UMASS CMPSCI Tech Report
TR98-008, Feb. 1998.
The URL is:
ftp://gaia.cs.umass.edu/pub/Padhye-Firoiu98:TCP-throughput.ps.Z
You can find a compilation of other papers on the subject on the TCP
Friendly website:
http://www.psc.edu/networking/tcp_friendly.html
You will need to do the same calculation Fred showed to convert BER to packet
loss rates. I think you will find that BER at 1E-8 is the highest acceptable,
and for fast links you might find 1E-12 to be desirable.
--Jamshid
This archive was generated by hypermail 2b29 : Mon Feb 14 2000 - 16:14:54 EST