[altq 44] Re: RED parameters

[Kenjiro Cho <kjc@csl.sony.co.jp>]

CCed to the altq mailing-list.

>>        I saw the difference in RED paper and FREEBSD code for
>>	  dropping probability calculation. Can you please explain it ? 

>>        RED paper says
>>        Pa = Pb/ (1 - count*Pb)
>>        Pb = MAXp (avg - MINth)/(MAXth - MINth)

>>        BSD code does this
>>        for denominator
>>        d = (2* (TH_MAX - TH_MIN) / MAXp)

>>        so why we are having this extra 2 ?       

This corresponds to the "wait" parameter set to "true" in the ns
simulator.  In this mode, the drop probability starts from Pb/2
(instead of Pb) and provides larger spacing between drops.
The similar effect can be obtained by decreasing MAXp to MAXp/2.

>>        Similarly, 
>>        Pa = Pb/ (2 - count*Pb)

>>        Is it that we have arrived at this equation after simulations ?

I believe so.  At least, it is the current default in ns.

>>        also inverse of maximum prob. is set at 10. is it arrived at
>>	  to use Sally's recommended values (1/ 30) ? 

The current default in ns is MAXp = 0.1.
As I recall, Sally said that it works better with short lived flows.

These parameters are briefly described in
"Ns Simulator Tests for Random Early Detection (RED) Queue Management"
http://www.aciri.org/floyd/red.html

>>         One thing more, RED i understand is also implmented on
>>	   RTOS. Do we have any performance figures for that? Even if you give
>>	   pointers it is sufficient. 

RTOS == Real Time Operating Systems?
The performance should not be different from a FreeBSD box.

>>         Thanks in advance.
>>                           -Samvid

-Kenjiro