RED and WRED in NS-3

[Javier Figueroa]

Hi everybody.

When applying QoS to incoming classified packets, the router must now place the packets in different virtual queues according to their specific class. To do this, i would like to implement red and wred queue. I can see that with NS-3 i could implement RED and set the min and max threshold but, how does it calcule the loss probability? I want to be capable of playing with this value because i would like to trigger different QoS policies to the same flows depending on the level of deprecation i have.

My other question would be, is WRED supported on NS-3? I would like to have in the same queue (lets say, class A high priority queue) packets with different priorities, meaning different packet loss probabilities.

Thank you!

[Tom Henderson]

On 11/30/2014 04:07 AM, Javier Figueroa wrote:
> Hi everybody.
>
> When applying QoS to incoming classified packets, the router must now
> place the packets in different virtual queues according to their
> specific class. To do this, i would like to implement red and wred
> queue. I can see that with NS-3 i could implement RED and set the min
> and max threshold but, how does it calcule the loss probability? I want
> to be capable of playing with this value because i would like to trigger
> different QoS policies to the same flows depending on the level of
> deprecation i have.

There is a RED queue model found in src/network/model/red-queue.{cc,h}.
It was ported from ns-2 (from the RED found in the queue/ directory of
ns-2). It does not presently support ECN marking. Some documentation
can be found here.

http://www.nsnam.org/docs/models/html/queue.html#random-early-detection

>
> My other question would be, is WRED supported on NS-3? I would like to
> have in the same queue (lets say, class A high priority queue) packets
> with different priorities, meaning different packet loss probabilities.

WRED is not supported. We are planning to do some refactoring to more
easily enable access to the IP header to support classification based on
IP/transport headers. This is discussed in this recent thread, starting
here:
http://mailman.isi.edu/pipermail/ns-developers/2014-November/012398.html

- Tom