[pox-dev] send packet from controller

Silvia Fichera fichera.sil at gmail.com
Sat Jul 20 15:57:35 PDT 2013


Thanks for the summary.
I was planning to use the packet_out but I am missing the connection object
(I saw this is usually taken from an event). Can I create a connection
object (to link controller and a switch) starting the communication from
the controller?


2013/7/20 Murphy McCauley <murphy.mccauley at gmail.com>

> Let me try restating.
>
> Imagine you have a hardware switch.  It has some ports which connect it to
> other switches and hosts and stuff -- it makes a network (call this the
> "data network" for lack of a better term).
>
> So if it's an OpenFlow switch, where is the controller in this picture?
>
> One type configuration, the controller can be anywhere on the data
> network.  It communicates with the switch over IP, so why not?  You have
> OpenFlow mixed with normal traffic all over your network.  (This is often
> called "in band control")
>
> In another type of configuration (out of band control), you set aside one
> port on each switch as "special", and use this port to connect to the
> controller.  The special port is *not* a part of the data network.  You
> might say it's part of a control or management network.  No normal
> forwarding ever takes place over this port -- *only* OpenFlow.
>
>
> If your setup looks like the first configuration, a controller can easily
> send arbitrary traffic over the data network using plain old socket
> programming.  But if your configuration is similar to the second option,
> there's no direct way for the controller to send to or receive from the
> data network.  The only way it can do it is over OpenFlow -- by instructing
> one of the switches to send data (via a packet out) and having the switch
> send data to the controller (via packet in).
>
> Of course, even in the second configuration, you could run a cable between
> a second interface on the controller and a normal port on the switch.  That
> way you have two cables between switch and controller -- one on the
> management network and one on the data network.
>
>
> Hope that helps.
>
> -- Murphy
>
> On Jul 20, 2013, at 8:49 AM, Silvia Fichera wrote:
>
> >  In which case, there's no way to send except via a datapath or a host
> which actually is.
> Can you clarify this sentence?
>
> Am I following the right way to send packets from the controller to hosts
> through switches, so using my network, or do I miss something?
>
>
> 2013/7/20 Murphy McCauley <murphy.mccauley at gmail.com>
>
>> If I'm understanding correctly, the problem is that the controller isn't
>> necessarily *on* the data network.  In Mininet, for example, it is often
>> the case that the controller and the datapaths are linked essentially by a
>> separate management network, and this is not an unusual case in the real
>> world either.  In which case, there's no way to send except via a datapath
>> or a host which actually is.
>>
>> Hope that helps.
>>
>> -- Murphy
>>
>> On Jul 19, 2013, at 3:39 PM, Silvia Fichera wrote:
>>
>> Hello,
>> I am having trouble sending packets from l3_learning controller to host.
>> I would like to send UDP packets but if I try to use normal socket from
>> controller I see no traffic (I minotired it with wireshark on all switches)
>> unless I send to itself (127.0.0.1).  I was also trying to make a switch
>> sending the controller generated packet in this way:
>>
>> http://lists.noxrepo.org/pipermail/pox-dev-noxrepo.org/2012-October/000281.html
>>
>> Although I have no traffic too.
>> I guess the problem is that I am missing something like
>>
>>  self.connection.send(msg)
>>
>> but I don't have any datapath connection with any switch since I want to
>> start the communication from the controller.
>> Is there an easier way to send these udp packets?
>>
>> thanks
>>
>> --
>> Silvia Fichera
>>
>>
>>
>
>
> --
> Silvia Fichera
>
>
>


-- 
Silvia Fichera
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.noxrepo.org/pipermail/pox-dev-noxrepo.org/attachments/20130721/041d679e/attachment-0002.htm>


More information about the pox-dev mailing list