Re: [ns] change flow id at intermediate node

[Tarik Alj <aljtarik@inrs-telecom.uquebec.ca>]

>Date: Mon, 16 Oct 2000 12:45:53 -0200 (EDT)
>From: Carlos Alberto Kamienski <cak@cin.ufpe.br>
>X-Sender: cak@paulista
>To: Tarik Alj <aljtarik@inrs-telecom.uquebec.ca>
>cc: ratul@aciri.org, ns-users@isi.edu
>Subject: Re: [ns] change flow id at intermediate node
>MIME-Version: 1.0
>
>> Since the queue is contained in a Link object, strictly speaking *action* 
would 
>> not take place at the *intermediate* node anymore. That is, if you were to 
>> replace the queueing scheme.
>
>This is because in ns queues belong to links and not to nodes, but this is
>an idiosyncrasy of ns.

not really, usally there is an output queue per *physical* port (correct me if 
I'm wrong). Besides a queue per se is not an object that should mark/remark 
packet; I think we both agree on that.  

>
>I think the best place to put this function (that classifies e change
>fid_) is before the classifier.
>Sean Murphy's diffserv implementation does it.
>
>If you want to use a queue type that is based on the fid_ (like CBQ) you
>should certainly change fid_ value before it is sent to that queue.
>In addtion, if you want to choose an output line (link) based on the fid_,
>you should change it before the classfier.

One could want to classify, then choose wheter or not to remark the packet then 
send it to whatever queueing engine that will make use of this info. No one 
prevents you from cascading classifiers/filters/markers. 

>
>Carlos
>
>


Tarik