I haven't really thought about it, but I think all of the basic "learning" 
forwarding components in POX have this same restriction.  Getting around it 
would require implementing something additional (e.g., using one of the 
strategies I outlined earlier).

-- Murphy

On Sep 12, 2013, at 1:36 AM, Dharani Kunnuru <[email protected]> wrote:

> On Thu, Sep 12, 2013 at 1:21 PM, Murphy McCauley <[email protected]> 
> wrote:
> In the future, please consider not dropping the list.
> 
> 
> None of the components which ship with POX are intended for this use case, 
> but it can be done in several ways.  For example...
> 
> One straightforward approach would be to implement STP-over-OpenFlow.  
> Another is to have controllers communicate between each other to jointly 
> discover the whole network's topology (this is the approach taken by ONOS).  
> If you don't do controller assignment at random and instead have multiple 
> "islands", then you can do a two-level discovery (discovery within islands 
> and discovery between islands).
> 
> There's a whole space of designs, and I'm not sure that there are any clear 
> universal winners.
> 
> -- Murphy
> 
> On Sep 12, 2013, at 12:21 AM, Dharani Kunnuru <[email protected]> 
> wrote:
> 
>> Thank you for your reply..
>> 
>> Yes it is working for single controller case. I have tried with 3 modules 
>> forwarding.l2_learning, openflow.spanning_tree, openflow.dicovery.  
>> Forwarding is possible incase of loops also. 
>> 
>> The same i tried with multiple controllers. It is not working. 
>> 
>> I have created 4 switches as follows: 
>> 
>> s1-------s2
>> |           |
>> |           |
>> s3-------s4
>> 
>> connection between s3---s2  and s1---s4 are also there. 
>> 
>> I have assigned switches randomly under two controllers c1 & c2 . 
>> 
>> In this scenario, no two hosts are communicating. So , is there any way to 
>> make this work.
>> 
>> please let me know the solution.
>> 
>> 
>> On Thu, Sep 12, 2013 at 6:48 AM, Murphy McCauley <[email protected]> 
>> wrote:
>> Most (or all?) of the forwarding components POX ships with will cause 
>> broadcast/flood storms when used on topologies with loops by themselves.  
>> This isn't a property of POX, but of the specific forwarding components.  It 
>> can be resolved in multiple ways.  One way is to disable flooding on some 
>> switch ports such that the flood-enabled ports form a spanning tree.  The 
>> openflow.spanning_tree POX component implements this method in a way which 
>> is at least partially agnostic to the forwarding component.
>> 
>> So the short answer is yes.  If you, for example, run l2_learning and 
>> openflow.spanning_tree (the latter of which also requires 
>> openflow.discovery, if I recall correctly), forwarding should work even on 
>> topologies with loops.  See the POX manual wiki and probably the 
>> openflow.spanning_tree docstring for more on this.
>> 
>> -- Murphy
>> 
>> On Sep 11, 2013, at 11:02 AM, Dharani Kunnuru <[email protected]> 
>> wrote:
>> 
>>> Hello Everyone,
>>> 
>>> I want to use POX controller to distributed control plane ( Logically 
>>> centralized but physically distributed) . And the topology iam going to use 
>>> should support loops? So, I want to know whether POX supports loops or not?
>>> 
>>> I have tried with single controller, All hosts are not able to communicate 
>>> with each. I have read one post in this group
>>> 
>>> http://lists.noxrepo.org/pipermail/pox-dev-noxrepo.org/2013-July/000834.html
>>> 
>>> They mention some changes to switch to allow loops? So is it possible to 
>>> configure switch to allow loops?? 
>>> 
>>> Please some one reply me..
>>> 
>>> -- 
>>> K Dharani
>> 
>> 
>> 
>> 
>> -- 
>> K Dharani
> 
> 
> Sorry for breaking the list...
> 
> And thank you for your reply..
> 
> 
> 
> 
> 
> There is restriction in floodlight controller like only one link is allowed 
> between the Openflow and non-Openflow islands. 
> 
> Like this, any kind of restrictions in POX. 
> Please let me know
> -- 
> K Dharani

Reply via email to