Hi guys,

I solved the problem :)
Instead of inserting a pair of actions inside the edge switches, I inserted
only a OFPP_CONTROLLER output action.
When receiving notification that a packet is flowing by the first edge
switch, the controller sends an instruction to output the packet on correct
output port, through the *send_openflow_packet()* command.
In such a way the system in kept synchronized, and the system works
properly.

Have good holidays

2011/8/12 andrea simeoni <andreasimeon...@gmail.com>

> Hi guys,
>
> I'm writing a simple NOC C++ controller that estimates the end-to-end delay
> of a UDP flow in a OF network.
> To make such estimation I do the following:
>
> When the flow to be monitored is received, I install a route using the
> routeinstaller component, with a pair of actions on the edge nodes of the
> path (OFPP_CONTROLLER and normal forwarding on path); while for all switches
> in the middle of the path I setup  a normal forwarding action.
> In such a way I'm able to understand when a packet enters and exits the
> network, an can estimate the delay (asking also the switchrtt component to
> accomodate the rtt).
>
>
>  When calling the routeinstaller install_route(..) function, I always get
> an OF malformed packet exception. I get the same error even if running the
> simple_routing component, that calls the same routeinstaller API. However
> despite such error notification, I can see that the route is correctly
> installed, by hand inspection of  flow tables.
>
> My problem is that once the route is installed, given the same UDP flow in
> input, the system behaves in a non deterministic way:
>
>     > Sometimes I get the first packet-in notification from the first
> switch in the path, and sometimes from the last.
>     > Sometimes the number of outgoing packets is equal to the number of
> ingoing packets + 1 (weird).
>     > Sometimes a big set of packets is first notified by the outgoing
> switch (very bad behaviour).
>
>
>
> Any suggestions?
>
> -Best regards-
>
> --
> Andrea Simeoni
>



-- 
Andrea Simeoni
_______________________________________________
nox-dev mailing list
nox-dev@noxrepo.org
http://noxrepo.org/mailman/listinfo/nox-dev

Reply via email to