[altq 234] Re: Diffserv and the RIO dropper

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

A class can be per micro-flow or per BA (behavior aggregate) depending
on classifiers assigned to the class.
In a diffserv network, edge routers have MF (multi-field) classifier
and mark an AF drop precedence value to a packet according to per
customer contract.  On the other hand, core routers have only BA
classifiers and drop packets according to their AF drop precedence.

A riometer in altq-1.2 is bound to a CBQ class.
So, if you want to do micro-flow based marking and behavior aggregate
based dropping with altq-1.2, you need 2 hops.
In altq-2.0, traffic conditioners are separated from queueing
disciplines so that you can do it in a single router.

-Kenjiro

Stewart Fallis wrote:
> I have a question regarding the RIO dropper within AltQ. 
> 
> class cbq fxp0 af-1 def_class borrow pbandwidth 1 rio 
>       filter fxp0 af-1 0 0 10.0.1.0 netmask 0xffffff00 0 0 
>       filter fxp0 af-1 0 4002 0 0 0 
>       riometer fxp0 af-1 2M 32K 0x40 meteronly class
> cbq fxp0 af-4 def_class borrow pbandwidth 1 rio 
>       filter fxp0 af-4 0 0 10.0.2.0 netmask 0xffffff00 0 0 
>       riometer fxp0 af-4 2M 32K 0xa0 meteronly 
> 
> In the above example traffic destined for the 10.0.1.0/24 network and all
> traffic sent from port 4002 will be classified into the af-1 class.
> 
> My question is since the RIO meter is per class (I am assuming it is) then it
> cannot determine what is in and out traffic per flow. My interpretation of
> diffserv was that in/out dropping was done per flow so as "greedy" flows would
> be dropped in preference to conforming flows. This does not seem to have been
> implemented in AltQ 1.2 (has it been rectified in 2.0?)
> 
> Have I interpreted that use of the riometer option correctly?
> 
> -- 
> Stewart Fallis <stewart@mirror.bt.co.uk> 
> Connectionless Networking 
> Advanced Communications Research
>