[e2e] RED parameters

[Saverio Mascolo]

I agree with you. We found some results reported in

http://www-ictserv.poliba.it/mascolo/papers/Easy_red.pdf

-Saverio

> Hi all,
>
> The recommendations for setting RED paramters at
>
> "http://www.aciri.org/floyd/REDparameters.txt"
>
> suggest that
> a) max_th should be set to three times min_th and
> b) min_th should be set to 5 packets.
>
> However, shouldn't the value of max_th be based on maximum buffer size
> too? For example, if the buffer size is large, max_th should also be
> large, otherwise it might result in a high packet drop rate. A number of
> papers on RED (in the simulation results) mention max_th and min_th
> without mentioning the buffer size. Do max_th and min_th make sense if the
> buffer size is not specified?
>
> Also as far as reducing the drop rate is concerned, some simple
> simulations in ns show that max_th set close to the buffer size gives the
> least drop rate. (for example, the default ns settings use B=50 packets,
> max_th=15 and min_th=5. Simply changing max_th to 45-50 results in a
> drastic improvement in drop rate in a congested scenario.) What is the
> motivation behind setting max_th to 15 when the buffer size is 50?
>
> Also, for low buffer sizes, the performance of Drop-Tail is better than
> RED in terms of drop rate. What advantage(s) does RED give in such
> scenarios? The reduction is average delay might not be significant
> compared to the increase in the packet drop rate if RED is used.
>
> -Rahul.
>