[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Fair Queueing




Taroon,

Your understanding of how buffer management is done in the FQ
implementation in ns sounds correct.

As an alternative, there is a Deficit Round Robin implementation
in ns which has a global pool of buffers shared among different
flows.  I believe its drop policy is to drop from the queue of
the flow occupying the most buffer space when the buffer is
full.

As I recall, the FQ and DRR objects are architected very differently.
Off hand, I'm not sure whether it would be easier to add your
desired buffer management to the FQ implementation, or to change
the scheduling behavior of the DRR implementation to do Fair Queueing.

Lee

Date: Thu, 17 Jun 1999 12:52:17 -0400 (EDT)
From: Taroon Mandhana <[email protected]>
To: [email protected]
Subject: Fair Queueing

Hi,

 I am using FQ in ns. What I could understand, for FQ by default, in the
link, the head_ points to a FQ classifier_ which has pointers to various
DropTail queues_ (corresponding to each flow) and each of these DropTail
queues_ point to a FQ object(which maintains the finish tags etc etc). So
when an enque happens on one of the flows(it happens on one of the
DropTail queues), it checks for the qlim_(on the droptail queue for that
flow)  to decide about dropping. But, this qlim_ is seperate for each flow
queue.

 I am trying to simulate a router with fixed number of buffers, hence I
want to impose qlim_ for global queue and not individual flows. Is that
feature already available in ns?? 

Thanks,
Taroon