[nox-dev] The LLDP problem when using Extreme OpenFlow-enabled switch
Hi, Recently, we use the discovery and lavi applications in NOX. The experiment environment contains two different switches, one is PC w/ NetFPGA and the other is Extreme OpenFlow enabled-Switch. After starting the discovery app, we got the complaint messages on the console. The messages are showed below: 00103|pyrt|ERR:unable to invoke a Python event handler: Traceback (most recent call last): File ./nox/lib/util.py, line 116, in f event.total_len, buffer_id, packet) File ./nox/netapps/discovery/discovery.py, line 163, in lambda discovery.lldp_input_handler(self,dp,inport,reason,len,bid,packet), File ./nox/netapps/discovery/discovery.py, line 251, in lldp_input_handler assert (packet.type == ethernet.LLDP_TYPE) AssertionError After trying to debug these two days and check the ethernet frame, we print received packet and find some hints from discovery.py [(Extreme Networks):51:a9:8501:23:20:00:00:01:VLAN](vlanid=10: pcp=0)[LLDP]chassis ID:00:04:96:51:a9:85port ID:00:01ttl:120tlv end [(Nicira Network):a3:71:e301:23:20:00:00:01:LLDP]chassis ID:00:23:20:a3:71:e3port ID:00:02ttl:120tlv end The Extreme Switch will pack the VLAN info into the OFP packet, so we modify a little parts of ethernet.py. This code just skip the VLAN part (4 bytes) of ethernet frame. The attachment is our modified code. Although it works, we are not sure doing this is right or not. We are not the python expert and maybe you guys could provide better way for this. Thanks, Jen-Wei ethernet.patch Description: Binary data ___ nox-dev mailing list nox-dev@noxrepo.org http://noxrepo.org/mailman/listinfo/nox-dev
Re: [nox-dev] The LLDP problem when using Extreme OpenFlow-enabled switch
Hi, Thanks Murphy. It works and I also think your suggestion is a good way. And thank again for how to use of packet.next. I think this is useful when the packet also contains another ethernet type inside. Hope my thought is right. Jen-Wei On Thu, May 26, 2011 at 5:36 PM, Murphy McCauley jam...@nau.edu wrote: Hmm... I don't immediately know what the right thing to do in this situation is. But if it's to ignore the vlan (which doesn't immediately seem completely unreasonable), I think maybe the right approach is to replace that assert at the start of lldp_input_handler() with something like: if packet.type == ethernet.VLAN_TYPE: if not packet.next: lg.error(lldp_input_handler could not parse vlan frame) return packet = packet.next assert (packet.eth_type == ethernet.LLDP_TYPE) else: assert (packet.type == ethernet.LLDP_TYPE) That's 100% untested, so if you wanted to give it a try, please let me know how it turns out. :) Anyone want to comment on whether that's the right approach? -- Murphy On Thursday, May 26, 2011 01:47:00 AM Jen-Wei Hu wrote: Hi, Recently, we use the discovery and lavi applications in NOX. The experiment environment contains two different switches, one is PC w/ NetFPGA and the other is Extreme OpenFlow enabled-Switch. After starting the discovery app, we got the complaint messages on the console. The messages are showed below: 00103|pyrt|ERR:unable to invoke a Python event handler: Traceback (most recent call last): File ./nox/lib/util.py, line 116, in f event.total_len, buffer_id, packet) File ./nox/netapps/discovery/discovery.py, line 163, in lambda discovery.lldp_input_handler(self,dp,inport,reason,len,bid,packet), File ./nox/netapps/discovery/discovery.py, line 251, in lldp_input_handler assert (packet.type == ethernet.LLDP_TYPE) AssertionError After trying to debug these two days and check the ethernet frame, we print received packet and find some hints from discovery.py [(Extreme Networks):51:a9:8501:23:20:00:00:01:VLAN](vlanid=10: pcp=0)[LLDP]chassis ID:00:04:96:51:a9:85port ID:00:01ttl:120tlv end [(Nicira Network):a3:71:e301:23:20:00:00:01:LLDP]chassis ID:00:23:20:a3:71:e3port ID:00:02ttl:120tlv end The Extreme Switch will pack the VLAN info into the OFP packet, so we modify a little parts of ethernet.py. This code just skip the VLAN part (4 bytes) of ethernet frame. The attachment is our modified code. Although it works, we are not sure doing this is right or not. We are not the python expert and maybe you guys could provide better way for this. Thanks, Jen-Wei ___ nox-dev mailing list nox-dev@noxrepo.org http://noxrepo.org/mailman/listinfo/nox-dev
Re: [nox-dev] Have any application in NOX supports multicast?
Dear Murphy KK, Thanks for your kindly reply and help. Just like KK mentioned, we might need an IGMP multicast. So, in this point, we must do something on NOX. That is, we need to implement a data structure that do the same thing as Group table for recording which ports have IGMP request. Is is right, please clear me! Thanks Jen-Wei On Mon, Mar 28, 2011 at 9:22 AM, kk yap yap...@stanford.edu wrote: Hi Jen-Wei, OpenFlow 1.1 (implemented) is released, so it is quite final for me. Also, if you are not constrained to using the group semantics (in OpenFlow 1.1), you can do multicast by listing all the ports you want to multicast to in the flow_mod command. If you are looking for IGMP multicast support, then you are looking for someone to do some work. Regards KK On 27 March 2011 17:41, James Murphy McCauley jam...@nau.edu wrote: So this is certainly not an official statement of any sort, but I would not expect to see OpenFlow 1.1 support in NOX for a while. I believe there's some consensus that supporting both 1.0 and 1.1 well will require some nontrivial API changes (and we're due for some API changes anyway), and there will be a focus on getting some of that right rather than incrementally updating NOX as has been done for previous OpenFlow versions. And before we even have 1.1 compatibility, I really would not hazard a guess about what we'd be releasing supporting multicast. Sorry! Hopefully we'll have more of an idea about this in the coming months, as well as more actual OpenFlow 1.1 implementations. (Also, as far as my limited understanding goes, OpenFlow 1.1 is still not final...) -- Murphy On Sat, 2011-03-26 at 11:33 +0800, Jen-Wei Hu wrote: Hi all, We are glad to hear the version 1.1 of OpenFlow is released because this version create a Group table for supporting multicast. Although we use the NOX that support OpenFlow 1.0, would any plan to develop an application or example which supports multicast? Or give us some advices to do it? Thank you! Bests, Jen-Wei ___ 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 ___ nox-dev mailing list nox-dev@noxrepo.org http://noxrepo.org/mailman/listinfo/nox-dev
Re: [nox-dev] Have any application in NOX supports multicast?
Dear Kyriakos, Thanks for your quickly help. Last week, I found in lib/packet have some files, such as arp or icmp. That time, I think may not follow them to do similar thing. But after your explanation, I think we are wrong and must to do it. The other thing is we also record which ports contain vedio streaming, is it right? Thanks Jen-Wei On Thu, Mar 31, 2011 at 2:29 PM, Kyriakos Zarifis kyr.zari...@gmail.comwrote: Hi, although that would take some effort, it is perfectly doable to hold that information on your component. You wouldn't have to use v1.1 tables or anything like that, all the state would be on your controller. While there are several packet types of packets included in the packet parsing library of NOX, IGMP messages are unfortunately one of them. So in order to extract the Join/Leave Group information, you would have to parse them manually (or add a relevant igmp.py in lib/packet/, following the paradigm of the other packet classes. such an addition would be welcome). On Wed, Mar 30, 2011 at 11:20 PM, Jen-Wei Hu jenwe...@gmail.com wrote: Dear Murphy KK, Thanks for your kindly reply and help. Just like KK mentioned, we might need an IGMP multicast. So, in this point, we must do something on NOX. That is, we need to implement a data structure that do the same thing as Group table for recording which ports have IGMP request. Is is right, please clear me! Thanks Jen-Wei On Mon, Mar 28, 2011 at 9:22 AM, kk yap yap...@stanford.edu wrote: Hi Jen-Wei, OpenFlow 1.1 (implemented) is released, so it is quite final for me. Also, if you are not constrained to using the group semantics (in OpenFlow 1.1), you can do multicast by listing all the ports you want to multicast to in the flow_mod command. If you are looking for IGMP multicast support, then you are looking for someone to do some work. Regards KK On 27 March 2011 17:41, James Murphy McCauley jam...@nau.edu wrote: So this is certainly not an official statement of any sort, but I would not expect to see OpenFlow 1.1 support in NOX for a while. I believe there's some consensus that supporting both 1.0 and 1.1 well will require some nontrivial API changes (and we're due for some API changes anyway), and there will be a focus on getting some of that right rather than incrementally updating NOX as has been done for previous OpenFlow versions. And before we even have 1.1 compatibility, I really would not hazard a guess about what we'd be releasing supporting multicast. Sorry! Hopefully we'll have more of an idea about this in the coming months, as well as more actual OpenFlow 1.1 implementations. (Also, as far as my limited understanding goes, OpenFlow 1.1 is still not final...) -- Murphy On Sat, 2011-03-26 at 11:33 +0800, Jen-Wei Hu wrote: Hi all, We are glad to hear the version 1.1 of OpenFlow is released because this version create a Group table for supporting multicast. Although we use the NOX that support OpenFlow 1.0, would any plan to develop an application or example which supports multicast? Or give us some advices to do it? Thank you! Bests, Jen-Wei ___ 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 ___ 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
Re: [nox-dev] Cycles
Hi, Niky Thanks for your detail information. Currently, we also have same confusion. We also find there is a basic spanning-tree on Openflow Wiki ( http://www.openflowswitch.org/wk/index.php/Basic_Spanning_Tree). Is it included into NOX's apps? Does this spanning-tree app do the same thing which you described? Thanks, Jen-Wei On Wed, Nov 17, 2010 at 1:37 PM, Niky Riga nr...@bbn.com wrote: The problem might be triggered by broadcast packets. When the controller receives a packet that has as destination address the bcast MAC address (ff:ff:ff:ff:ff:ff), an ARP packet for example, it will send the packet to the special outport OFPP_FLOOD. OFPP_FLOOD is translated at the switch to all ports other than the input port. In your topology if End1 sends an ARP request packet then Openflow1 will send a copy to Openflow2 and Openflow3, both of which are going to send a copy to Openflow4. When Openflow4 receives a packet from Openflow2(3) then it will send a copy to End2 and to Openflow3(2) respectively. Openflow2(3) will give the packet back to Openflow1 and the loop starts again. The correct solution would be to implement a minimum spanning tree on the topology to disseminate broadcast packets. An easy way out, which I have used before, is to set the NO_FLOOD flag in some ports to break the loop. For example you can choose that Openflow1 will forward bcast packets only to Openflow2, and set the NO_FLOOD flag on the port that connects Openflow1 to Openflow3. This flag will basically exclude this port when the action specifies as outport the OFPP_FLOOD port. Note that if you are using a learning switch module then the loops will also cause multiple copies of the same packet to be delivered to the destination even for unicast packets. This happens because until the controller learns the mapping between mac addresses and ports, it will flood packets. Hope this helps, Niky Kyriakos Zarifis wrote: Are you running 'discovery'? If so, the events you see could be the LLDP packets sent by that component. On Tue, Nov 16, 2010 at 5:55 PM, Rohit Manohar rdman...@ncsu.edumailto: rdman...@ncsu.edu wrote: I am trying to build the following topology in Nox: Openflow2 -- - End1 -- Openflow1 Openflow4 End2 - - - - Openflow3 When I register Openflow4 with NOX, some sort of traffic loop is getting created. NOX is continuously registering packet-in events even when I haven't set up any flows. Are cycles not supported by NOX? If yes, do I need to indicate it in the .conf file? Regards, -- Rohit Manohar Graduate Student North Carolina State University Raleigh, US. ___ nox-dev mailing list nox-dev@noxrepo.org mailto:nox-dev@noxrepo.org http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org ___ nox-dev mailing list nox-dev@noxrepo.org http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org ___ nox-dev mailing list nox-dev@noxrepo.org http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org ___ nox-dev mailing list nox-dev@noxrepo.org http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org
Re: [nox-dev] The conversation between two NOXs
Hi martin, Thanks again! We still ask some questions inline. On Fri, Mar 26, 2010 at 9:56 AM, Martin Casado cas...@nicira.com wrote: Inline: Dear martin, Thanks for your reply. We just think when the network environment contains more than 2 NOXs, what is the behavior between them? So, NOX do not do the load balance currently? Correct, Nox does not support multiple instances working together. For that, you will have to extend Nox. If we want to modify, which source code is the good starting point? By the way, We still have some questions which need to your help. 1) What is the process that NOX want to build its controlled OpenFlow switches topology? NOX will send an event to notify all controlled OpenFlow switches, then all OpenFlow switches will send LLDP out from their owned ports to all connected switches? Right. Nox sends out LLDP packets from every switch port and then listens for incoming LLDP packets on neighbor switches to determine the topology. This means that the LLDP packets are sent out from NOX? When LLDP packets arrived on every switches which connect to NOX, every switch will redirect LLDP packets to its connected switches? 2) The discoveried topology will be stored on NOX? Correct. 3) Support we have a network topology, which showed below, these OpenFlow switches are all controlled by one NOX. What do the NOX write into the flowtables of these OpenFlow switches when user ping host2 from host1? host1 -- ofsw1 --- ofsw2 | | | | ofsw3 --- ofsw4 -- host2 Nox doesn't support topologies with loops (though there are a number of spanning tree implementations which have been ported to it) so this would not work. However, if there wasn't a loop, then nox would set up flows for the ARP and ICMP packets in both directions. We assume to remove the link between ofsw1 and ofsw2 in the above topology (no loop topology). NOX knows that ofsw1 connect to ofsw3, ofsw3 connect to ofsw4 by LLDP packets? If it is correct, will NOX write ofsw3's information into ofsw1, such as which port of ofsw1 connects to ofsw3 port? There will have the similar information on ofsw3 and ofsw4? Bests, JW .martin We really appreciate to all the help you gave us. Best, JW On Fri, Mar 26, 2010 at 5:55 AM, Martin Casado cas...@nicira.commailto: cas...@nicira.com wrote: Hi JW, Nox was not built to support coordination between two instances. Why don't control both switches from the same Nox process? .martin Dear all, We just join the world of OpenFlow and NOX lately. Everything for us is new and also doesn't know how to start. Currently, we have built the enviroment which contains 1 NOX (nox1), 1 OpenFlow switch with NetFPGA (ofsw1), and 2 hosts which connect to ofsw1. After installing, we got some quesions about it. Could someone could help us? (current) nox1 -- ofsw1 |--- host1 |--- host2 (suppose) nox1 -- ofsw1 - ofsw2 -- nox2 |--- host1 |--- host2 1. Suppose we add 1 OpenFlow switch (ofsw2) and 1 NOX (nox2), let this ofsw2 connects to nox2 and sw1. The LLDP will let nox1 discover ofsw1, how let nox1 also discovery ofsw2? 2. Does nox1 could discover nox2? 3. The packages between ofsw1 and ofsw2 are using OpenFlow protocol? And how about nox1 and nox2? Thanks Bests, JW ___ nox-dev mailing list nox-dev@noxrepo.org mailto:nox-dev@noxrepo.org http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org ___ nox-dev mailing list nox-dev@noxrepo.org http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org
Re: [nox-dev] The conversation between two NOXs
Thanks for your quick and nice reply! JW On Fri, Mar 26, 2010 at 2:26 PM, Martin Casado cas...@nicira.com wrote: Inline: On Fri, Mar 26, 2010 at 9:56 AM, Martin Casado cas...@nicira.commailto: cas...@nicira.com wrote: Inline: Dear martin, Thanks for your reply. We just think when the network environment contains more than 2 NOXs, what is the behavior between them? So, NOX do not do the load balance currently? Correct, Nox does not support multiple instances working together. For that, you will have to extend Nox. If we want to modify, which source code is the good starting point? Oh boy, that depends entirely on what you want to do. To begin with, I would suggest building a simple client/server component using twisted in python in order to exchange state between controllers. By the way, We still have some questions which need to your help. 1) What is the process that NOX want to build its controlled OpenFlow switches topology? NOX will send an event to notify all controlled OpenFlow switches, then all OpenFlow switches will send LLDP out from their owned ports to all connected switches? Right. Nox sends out LLDP packets from every switch port and then listens for incoming LLDP packets on neighbor switches to determine the topology. This means that the LLDP packets are sent out from NOX? When LLDP packets arrived on every switches which connect to NOX, every switch will redirect LLDP packets to its connected switches? Yes, Nox sends LLDP packets out of every port on every switch. 2) The discoveried topology will be stored on NOX? Correct. 3) Support we have a network topology, which showed below, these OpenFlow switches are all controlled by one NOX. What do the NOX write into the flowtables of these OpenFlow switches when user ping host2 from host1? host1 -- ofsw1 --- ofsw2 | | | | ofsw3 --- ofsw4 -- host2 Nox doesn't support topologies with loops (though there are a number of spanning tree implementations which have been ported to it) so this would not work. However, if there wasn't a loop, then nox would set up flows for the ARP and ICMP packets in both directions. We assume to remove the link between ofsw1 and ofsw2 in the above topology (no loop topology). NOX knows that ofsw1 connect to ofsw3, ofsw3 connect to ofsw4 by LLDP packets? If it is correct, will NOX write ofsw3's information into ofsw1, such as which port of ofsw1 connects to ofsw3 port? There will have the similar information on ofsw3 and ofsw4? This depends on what applications are running. If you are running routing then yes, Nox discovers the topology via LLDP packets and stores the information in the topology component. So Nox knows, internally, that ofsw1 is connect to ofsw3, ofsw3 is connect to ofsw1 and ofsw4, and ofsw4 is connected to ofsw3 and ofsw2. .m Bests, JW .martin We really appreciate to all the help you gave us. Best, JW On Fri, Mar 26, 2010 at 5:55 AM, Martin Casado cas...@nicira.com mailto:cas...@nicira.com mailto:cas...@nicira.com mailto:cas...@nicira.com wrote: Hi JW, Nox was not built to support coordination between two instances. Why don't control both switches from the same Nox process? .martin Dear all, We just join the world of OpenFlow and NOX lately. Everything for us is new and also doesn't know how to start. Currently, we have built the enviroment which contains 1 NOX (nox1), 1 OpenFlow switch with NetFPGA (ofsw1), and 2 hosts which connect to ofsw1. After installing, we got some quesions about it. Could someone could help us? (current) nox1 -- ofsw1 |--- host1 |--- host2 (suppose) nox1 -- ofsw1 - ofsw2 -- nox2 |--- host1 |--- host2 1. Suppose we add 1 OpenFlow switch (ofsw2) and 1 NOX (nox2), let this ofsw2 connects to nox2 and sw1. The LLDP will let nox1 discover ofsw1, how let nox1 also discovery ofsw2? 2. Does nox1 could discover nox2? 3. The packages between ofsw1 and ofsw2 are using OpenFlow protocol? And how about nox1 and nox2? Thanks Bests, JW ___
Re: [nox-dev] The conversation between two NOXs
Dear martin, Thanks for your reply. We just think when the network environment contains more than 2 NOXs, what is the behavior between them? So, NOX do not do the load balance currently? By the way, We still have some questions which need to your help. 1) What is the process that NOX want to build its controlled OpenFlow switches topology? NOX will send an event to notify all controlled OpenFlow switches, then all OpenFlow switches will send LLDP out from their owned ports to all connected switches? 2) The discoveried topology will be stored on NOX? 3) Support we have a network topology, which showed below, these OpenFlow switches are all controlled by one NOX. What do the NOX write into the flowtables of these OpenFlow switches when user ping host2 from host1? host1 -- ofsw1 --- ofsw2 | | | | ofsw3 --- ofsw4 -- host2 We really appreciate to all the help you gave us. Best, JW On Fri, Mar 26, 2010 at 5:55 AM, Martin Casado cas...@nicira.com wrote: Hi JW, Nox was not built to support coordination between two instances. Why don't control both switches from the same Nox process? .martin Dear all, We just join the world of OpenFlow and NOX lately. Everything for us is new and also doesn't know how to start. Currently, we have built the enviroment which contains 1 NOX (nox1), 1 OpenFlow switch with NetFPGA (ofsw1), and 2 hosts which connect to ofsw1. After installing, we got some quesions about it. Could someone could help us? (current) nox1 -- ofsw1 |--- host1 |--- host2 (suppose) nox1 -- ofsw1 - ofsw2 -- nox2 |--- host1 |--- host2 1. Suppose we add 1 OpenFlow switch (ofsw2) and 1 NOX (nox2), let this ofsw2 connects to nox2 and sw1. The LLDP will let nox1 discover ofsw1, how let nox1 also discovery ofsw2? 2. Does nox1 could discover nox2? 3. The packages between ofsw1 and ofsw2 are using OpenFlow protocol? And how about nox1 and nox2? Thanks Bests, JW ___ nox-dev mailing list nox-dev@noxrepo.org http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org ___ nox-dev mailing list nox-dev@noxrepo.org http://noxrepo.org/mailman/listinfo/nox-dev_noxrepo.org
