[e2e] Re: queue averaging introduces delay

[Zhang Miao]

Hi, Lidia
    In my opinion, you need not use "adaptive" algorithm here. 
One easier way is to change the update method of RED. The 
original RED uses "packet arrival" to trigger the re-calculation
of marking/dropping probability. With different C, the rate of
packet arrival is different. So it shows that the q_w is not
applicable for various C. If we periodically trigger the 
re-calculation, say, every 0.001 second, then we can solve
this problem. This method has been mentioned in the paper of
Firoiu and Borden (INFOCOM 2000), Misra, etal. (SIGCOMM2000), 
and Hollot etal (INFOCOM 2001).

>Hi Wei Wu,
>
>> Maybe in RED, C*q_w should be a constant value, which can determine the
>> responsiveness of RED actually. So higher capacity C is, the less q_w should be...
>
>Actually adaptive RED proposes that qw = 1 - e^(-1/C)
>If we take C*qw to be a constant, say, C*qw = 1, then qw = 1/C.
>
>It turns out that these functions are quite close to each other for
>large C, say C>10. Then, it could be indeed more practical to use simply
>qw=1/C, for links of a reasonable speed (10 pkts/s = 120kbit/s for 1500
>byte pkts). If that is the case, then we could even think of adjusting
>qw dynamically for links that have variable capacity (e.g. the multiple
>queue case that was discussed earlier), since the calculation is quite
>simple. Would that make any sense?
>
>Lidia.

*****************************************************************
*    Zhang Miao                                                 *
*    Ph.D candidate,Department of Computer Science & Technology * 
*    Tsinghua University,Beijing,China(100084)                  *
*    Tel: (8610)-62785822                                       *
*    Email: zm at csnet1.cs.tsinghua.edu.cn                        *
*    Web: http://netlab.cs.tsinghua.edu.cn/~zm (domestic only ) *
*****************************************************************