Yes, that is possible. This is where in-band vs out-of-band control comes to
play: If the physical topology looks exactly like what you described then
control of sw1 can only been done in-band (meaning, the control packets are
using the same links as the data packets). In order for that to be possible,
there need to be some initial flow entries in the intermediate switches to
allow a remote switch to connect to the controller. I don't think there is a
standardized way to set up in-band control, but I am guessing a simple
version would be something like:

- s3 connects to c0
- the controller receives packet_in from s3, which originated in s2 while it
tried to connect
- c0 sets a flow on s3 that allows traffic from s2 to reach it
- at this point s2 and s3 are connected to the controller
- sw2 picks up packet from s1 and sends packet_in to c0

OpenvSwitch has a pretty sophisticated approach to in-band control. If
you're interested you can take a look here :
http://openvswitch.org/cgi-bin/gitweb.cgi?p=openvswitch;a=blob_plain;f=DESIGN
(look for "in-band")

On Thu, Oct 20, 2011 at 5:39 AM, scolfield <kscolfi...@gmail.com> wrote:

> Hi,
>
> Thanks for your reply.
> It's possible openflow messages across several switches until achieve the
> controller?
> Such as host -> sw1 -> sw2 -> sw3 -> c0 or in a reverse way in linear
> topology?
> Consider that the host packet need to achieve a server connected to sw3and 
> there are only
> one controller linked to one switch.
>
> --
> scolf
>
>
> On Thu, Oct 20, 2011 at 06:32, Kyriakos Zarifis <kyr.zari...@gmail.com>wrote:
>
>> Hi,
>>
>> pyswitch itself is really not aware of any topology. It only stores state
>> that is relevant to switches separately (i.e. it stores a mapping of mac
>> addresses to local ports for a switch).
>>
>> Now, whether a control packet (like a packet_in) will reach the controller
>> if it's not directly physically connected to it is another issue, and is
>> irrelevant to what NOX application (eg pyswitch) is running. Switches are
>> connected to the controller through a separate control channel, so as long
>> as this channel has been established (which is when the switch first
>> connects to NOX), then the packet_ins will find their way to NOX through
>> that (irrespective of dataplane connectivity, which is established by NOX
>> applications)
>>
>> (another thing to consider is whether the switch-controller connection is
>> in-bound/out-of-bound)
>>
>> Does this make sense?
>>
>> On Wed, Oct 19, 2011 at 5:20 PM, scolfield <kscolfi...@gmail.com> wrote:
>>
>>> Hi,
>>>
>>> The pyswitch example can learn topology of networks? Such as a
>>> corporation architecture when
>>> there are hierarchical switches interconnected one to another creating
>>> several "layers" of switches,
>>> at this case, the one openflow controller connected to two distinct
>>> switches, can learn which ports
>>> to send a ping packets, for example?  The packet in messages will be
>>> forwarded automatically through
>>> switches until reach a controller?
>>>
>>> --
>>> scolfield
>>>
>>> _______________________________________________
>>> nox-dev mailing list
>>> nox-dev@noxrepo.org
>>> http://noxrepo.org/mailman/listinfo/nox-dev
>>>
>>>
>>
>
_______________________________________________
nox-dev mailing list
nox-dev@noxrepo.org
http://noxrepo.org/mailman/listinfo/nox-dev

Reply via email to