On Aug 20, 2014, at 8:05 AM, 张伟 <zhang...@126.com> wrote:
> Thank you very much for your reply! See below.
>
>
>
>
>
> At 2014-08-20 02:48:33, "Murphy McCauley" <murphy.mccau...@gmail.com> wrote:
>
> On Aug 19, 2014, at 8:22 PM, 张伟 <zhang...@126.com> wrote:
>
>> Hi all,
>>
>> I want to know for the application eg l2_learning. If we run this component,
>> l2_learning is single thread or multi-thread?
>
> Single threaded.
>
>> My simple test:
>> def _handle_PacketIn (event):
>> packet = event.parsed
>> msg = of.ofp_flow_mod()
>> msg.match = of.ofp_match.from_packet(packet, event.port);
>> msg.buffer_id = event.ofp.buffer_id
>> log.info("buffer_id %i", msg.buffer_id);
>> msg.idle_timeout = 10
>> msg.hard_timeout = 30
>> msg.actions.append(of.ofp_action_output(port = 0))
>> msg.data = event.ofp
>> event.connection.send(msg)
>>
>> When I print out the buffer_id message, sometimes it is not sequence. I got
>> the result like that:
>> INFO:packet:(udp parse) warning UDP packet data shorter than UDP len: 96 <
>> 962
>> INFO:misc.test_flow_miss:buffer_id 0
>> INFO:openflow.of_01:[00-00-00-00-00-01 2] connected
>> INFO:packet:(udp parse) warning UDP packet data shorter than UDP len: 96 <
>> 962
>> INFO:misc.test_flow_miss:buffer_id 2
>>
>> This makes me very confused. At first, I thought just one single thread for
>> the app. app processed the requests according to FIFO mode. Can anybody give
>> some help and explain some reason for my result?
>
> The buffers come from the switches. How the switches decide on buffer IDs is
> not specified. Are you sure your switch is sending buffer IDs that increment
> by one? Have you snooped the control connection to confirm this?
>
> Another possibility is that a higher priority packet-in handler is getting
> the event first and eating it. If all you're running is l2_learning (and
> not, say, discovery) this shouldn't be the case, though.
>
> Yes, I am sure that the switch sends the buffer IDs that increment by one.
> Actually the buffer ID starts from 0. On my above component, from the
> log.info I can see that the buffer id starts from 0. However, on the switch
> side, when receiving the flow_mod message, the first buffer_id is always
> UINT32_MAX. On the controller side, I only run my own component. So this
> repsents other component sends the flow_mod message with buffer_id is
> UINT32_MAX. Do you think where can send this message?
The first flow_mod is probably the controller clearing the tables, which POX
does by default. The command should be delete.
To further examine things, capture the OpenFlow traffic with POX's
openflow.debug component and then look at it with the OpenFlow Wireshark
dissector. Check the buffer IDs in the packet-ins.
>> Another question:
>> In pox wiki, when we want to install a flow entry: if the tcp port is 8080,
>> why in python code, we do not need to do htons translation. However, the
>> switch side receives the port is network endian.
>
> Because it would be awful if you had to do it yourself and POX takes care of
> it for you.
> Can you help to point out where in pox do the translation?
Probably in a call to struct.pack() in libopenflow_01.ofp_match.pack().
> -- Murphy
