How is Controller implemented in NS3

[Wang Tao]

Dear,

I am interested in the OpenFlow supported in NS3.

It is interesting that the controller is internally built in NS3 unlike a external way in other platforms.

So I am wondering is there a cost for computing path for the switches when the controller receiving the messages (e.g., packet-in messages) from connected switches?

Wish for your reply for discussion. lol

Best,

Tao

[Tom Henderson]

On 11/18/2014 07:07 PM, Wang Tao wrote:
> Dear,
>
> I am interested in the OpenFlow supported in NS3.
>
> It is interesting that the controller is internally built in NS3 unlike
> a external way in other platforms.

Well, only some trivial controllers are built in; a Drop controller, and
a Learning bridge controller. People could instead write or port other
virtual controllers for this model.

An external controller could be possible (interacting with the switch in
real-time) but only if the controller can connect with a non-secure
(non-SSL) channel to the switch model.

More information here:

http://www.nsnam.org/docs/models/html/openflow-switch.html

>
> So I am wondering is there a cost for computing path for the switches
> when the controller receiving the messages (e.g., packet-in messages)
> from connected switches?

Sorry, I didn't understand this question; could you state it differently?

- Tom

[Wang Tao]

Thanks for your attention, Tom.

I will check the website you have offered.

BTW, the cost I mean is that when a packet received by a OpenFlow switch, there may not be matched flow table to take forwarding actions, so the switch will send the packet or other messages (e.g., packet-in message defined in OpenFlow protocol) to the controller. The controller then computes the path and install flow table to the switch. So what I want to know is that the computation may consume time (you know, in the real environment).

1)Does this time consumption (I call it the cost) exist in NS-3 platform?

2)If exists, how to measure it?

[Tom Henderson]

On 11/18/2014 09:48 PM, Wang Tao wrote:
> Thanks for your attention, Tom.
>
> I will check the website you have offered.
>
> BTW, the cost I mean is that when a packet received by a OpenFlow
> switch, there may not be matched flow table to take forwarding actions,
> so the switch will send the packet or other messages (e.g., packet-in
> message defined in OpenFlow protocol) to the controller. The controller
> then computes the path and install flow table to the switch. So what I
> want to know is that the computation may consume time (you know, in the
> real environment).
>
> 1)Does this time consumption (I call it the cost) exist in NS-3 platform?

No, in general, processing time is not modeled in the ns-3 models.

- Tom