Hello Viral,

You can let switches connect to a virtual IP address (aka, IP alias) and
configure your controllers to share this virtual IP address among each
other. In case of a controller failure, an active one can replace the
failed one by acquiring the released (orphan) virtual IP address. You can
achieve this mechanism using Linux-HA or your own custom software. (See
"Controlling a Software-Defined Network via Distributed Controllers" paper
in NEM Summit 2012
Proceedings<http://nem-summit.eu/files/2012/10/2012_NEM_Summit_Proceedings.pdf>for
a much more sophisticated usage of this trick in a multiple controller
environment.)

In addition to virtual IP addresses, depending on the OpenFlow protocol you
are using, you can specify multiple controller addresses in the switch
configurations. When a switch fails to connect to a controller, it will try
the next one in the list. (The questions like "How many controllers I can
provide?", "How does the switch choose which controller to connect?" are
totally vendor dependent, AFAIK.)

Best.



On Tue, Jan 22, 2013 at 7:35 AM, viral parmar <[email protected]>wrote:

> Hello,
> I have gone through some papers on Software Defined Network and openflow.
> And my interest arouse along the "controller placement problem" to maximize
> the fault tolerance in a Data Center.
>
> I am about to setup 30 virtual nodes at my inst. lab, but i am bit
> confused with the way of defining the fault tolerance.
>
> I mean what are the possible fault tolerance at a data center ?? Some i
> have seen like "power surge", "device or computer failure","overload" but i
> am not sure that i can test anyone of this in a virtual Data center at my
> inst.
>
> Any Suggestions???
>
> Thank you
>
> Regards
> Viral Parmar
>
>
> _______________________________________________
> openflow-discuss mailing list
> [email protected]
> https://mailman.stanford.edu/mailman/listinfo/openflow-discuss
>
>
_______________________________________________
openflow-discuss mailing list
[email protected]
https://mailman.stanford.edu/mailman/listinfo/openflow-discuss

Reply via email to