Re: [ovs-discuss] Get rid of immortal ARP
> On Sep 20, 2016, at 9:21 AM, Warsang wrote: > > Thank you for your answer. I am talking about adding a TTL like field. I have > visibility on the switches from start and the end hosts at first arp request > they send. So what you are suggesting is using an STP like protocol just on > ARP broadcast? If you want use a protocol, then I'd just use STP, since it's a standard and OVS supports it. However, if you have a high-level view of the network, you can program the OVS flows in such a way that you don't use STP and make better use of your links that might otherwise be turned off to prevent loops. --Justin ___ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
Re: [ovs-discuss] Get rid of immortal ARP
Thank you for your answer. I am talking about adding a TTL like field. I have visibility on the switches from start and the end hosts at first arp request they send. So what you are suggesting is using an STP like protocol just on ARP broadcast? -Warsang On 09/20/16 19:57, Justin Pettit wrote: On Sep 20, 2016, at 2:34 AM, Warsang wrote: Hello everyone, I am simulating a fat tree topology with ECMP. When one of my host sends an ARP request in broadcast, the packet is broadcasted through all switches which creates more and more traffic because I have no STP. My question is simple. What is the best way to get rid of these immortal packets? I thought of using something similar to a TTL header which would be decremented every time the packet hits a switch and dropped when it gets to zero however this method seems a bit too complicated to implement and I was wondering if there was a better, more simple way to do so using OVS? I'm not sure what a TTL would look like in ARP, since it doesn't have such a field, and an L2 switch wouldn't normally decrement one even if there were. Do you have enough visibility into the topology to program OVS to only forward broadcasts in such a way that they won't form loops? I've seen some SDN applications do this to make better use of their links than can be achieved with STP. --Justin ___ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
Re: [ovs-discuss] Get rid of immortal ARP
> On Sep 20, 2016, at 2:34 AM, Warsang wrote: > > Hello everyone, I am simulating a fat tree topology with ECMP. When one of my > host sends an ARP request in broadcast, the packet is broadcasted through all > switches which creates more and more traffic because I have no STP. My > question is simple. What is the best way to get rid of these immortal > packets? I thought of using something similar to a TTL header which would be > decremented every time the packet hits a switch and dropped when it gets to > zero however this method seems a bit too complicated to implement and I was > wondering if there was a better, more simple way to do so using OVS? I'm not sure what a TTL would look like in ARP, since it doesn't have such a field, and an L2 switch wouldn't normally decrement one even if there were. Do you have enough visibility into the topology to program OVS to only forward broadcasts in such a way that they won't form loops? I've seen some SDN applications do this to make better use of their links than can be achieved with STP. --Justin ___ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss
[ovs-discuss] Get rid of immortal ARP
Hello everyone, I am simulating a fat tree topology with ECMP. When one of my host sends an ARP request in broadcast, the packet is broadcasted through all switches which creates more and more traffic because I have no STP. My question is simple. What is the best way to get rid of these immortal packets? I thought of using something similar to a TTL header which would be decremented every time the packet hits a switch and dropped when it gets to zero however this method seems a bit too complicated to implement and I was wondering if there was a better, more simple way to do so using OVS? Thank you in advance. -Warsang ___ discuss mailing list discuss@openvswitch.org http://openvswitch.org/mailman/listinfo/discuss