Could you check and see... do you ever see any link or switch down events, or is it all just up events?
And do you ever have a run that appears to work entirely as expected? Also, have you tried POX's l2_multi module? I've never run it on a fat tree. -- Murphy On Aug 24, 2012, at 11:29 AM, Tatyana Yatskevich wrote: > Hello, > > I created a fat-tree topology in Mininet and try to use NOX controller. I am > using spanning_tree module to avoid loops while broadcasting and routing > module to forward traffic to known destinations. Spanning tree is contracted > correct but routing module works weird. When I do a ping "h1 ping -c1 h3", > "h1 ping -c1 h5", "... h6", "... h7", ".... h8", NOX finds the shortest path > as it is supposed to (h1-s23-s21-s24-h3, h1-s23-s21-s18-s25-s27-h5,.....). > But when "h1 ping -c1 h4", the traffic goes like this: > h1-s23-s21-s18-s25-s19-Controller-s22-s24-h4 which is obviously not the > shortest path and I see 2 ARP reply, 2 ICMP request, 2 ICMP reply messages on > the loopback interface in Wireshark. When Controller gets ICMP request > message from s23, it installs flow entries to s23, s21, s18, s25 and > instructs s23 to send the packet; when it reaches s19, it is sent to the > Controller again which installs flow entries to s19, s22, s24 and instructs > s19 to send the packet. The same happens with "h1 ping -c1 > h9,h10,h11,h12,h13,h14,h15,h16". Next run the situation may change: for > instance, "h1 ping -c1 h4' works as it is supposed to work(shortest path is > calculated) but "h1 ping -c1 h3" travels the longer path. > > Why does it work like that? > I will appreciate any help > Thank you very much > > Tatyana > > > <topology.png>
