[ns] Error Models for TCP performance evaluation over lossy links (fwd)

[Lloyd Wood <l.wood@eim.surrey.ac.uk>]

Forwarding this as relevant to ns.

<L.Wood@surrey.ac.uk>PGP<http://www.ee.surrey.ac.uk/Personal/L.Wood/>

---------- Forwarded message ----------
Date: Tue, 21 Nov 2000 15:13:47 -0800 (PST)
From: A. Abou-Zeid <hussein@lagrange.ee.washington.edu>
To: pilc@grc.nasa.gov
Cc: Marcelo Albuquerque <marcelo@lagrange.ee.washington.edu>,
     Prof. Sumit Roy <roy@lagrange.ee.washington.edu>
Subject: Error Models for TCP performance evaluation over lossy links



The attached note describes briefly our experience with regard to the
above mentioned topic, with pointers to more detailed information that
may be of interest to this list.

Most of the research involving TCP and lossy channels face a critical 
decision when choosing the most adequate error model to be used. One
should use caution  when opting between a discrete-time two-state error
model, where channel state  transitions occur synchronously with packet
transmissions, and a continuous-time two-state error model, where channel
state transitions occur independently from packet transmissions. In the
network simulator (http://www.isi.edu/nsnam), the  discrete-time two state
model corresponds to the former while the continuous-time model proposed
in (http://students.washington.edu/abouzeid/research.htm) corresponds to
the latter.

The discrete-time model has the advantage of allowing for a direct computation 
of relevant statistics such as average packet error rate and correlation between
errors. Expressing results such as throughput as a function of these parameters 
is quite standard in most research work. However, the model fails to be 
physically realistic since its state transitions are dependent on TCP dynamics. 
In some situations this translates into unreasonable patterns in channel state 
transition. The continuous-time model makes the channel state independent from 
TCP dynamics but makes it very difficult to obtain the most widely used error 
statistics, such as packet error rate. Please refer to 
http://students.washington.edu/marcelo/errormodels.html for a more 
detailed report on this issue.
 
Best regards,
Alhussein for Marcelo Albuquerque, Alhussein Abouzeid, Sumit Roy
Fundamentals of Networking Laboratory (FuNLab)
Department of Electrical Engineering
University of Washington, Seattle