Murphy,

I don't think that "link down" event is an issue even if there are some. If "h1 
ping -c1 h3" doesn't work correctly, it doesn't work correctly all the time and 
this can be changed only during next run. And if we suggest that it doesn't 
follow the shortest path because link s21-s24 is down then "h1 ping-c1 h4" 
would act the same way (instead, it follows the shortest path that goes through 
the link s21-s24).

Yes, I'm using spanning_tree module. Sometimes, I see some extra ICMP messages. 
Maybe here "link down" events have something to do with it :) because next 
pings(during the same run) give normal results. I will try to use your version 
of POX.

Thank you very much

Tatyana
________________________________
From: Murphy McCauley [[email protected]]
Sent: Sunday, August 26, 2012 10:04 PM
To: Tatyana Yatskevich
Cc: [email protected]
Subject: Re: [nox-dev] issues with routing module

On Aug 26, 2012, at 6:51 PM, Tatyana Yatskevich wrote:
I don't see any down events and I never had a completely correct run. I 
attached a snapshot of passing control messages during the ping (destination 
MAC address was already discovered). We can see that the controller installs 
flow entries to first 4 switches and then to the rest 3. There are some 
"Poisoning message" from authenticator module which I don't understand.

I actually don't remember if the down events will get printed on the console by 
default; it may make sense to write a small component that just prints link and 
switch down messages.

My reason for asking is that I have long suspected that there may be an issue 
where you don't get optimal paths after links go down, but have never looked 
into it really deeply (and at this point, I probably never will).

I tried POX controller and it looks like it works correctly so far. But I see a 
million of LLDP messages dropped.They don't seem to effect the network's 
working but they are disturbing cause they make a passing messages analysis 
more difficult.

Are you also running the spanning_tree component?  If not, you probably should. 
:)  You also might get somewhat better results with the version in the 
github.com/MurphyMc/pox<http://github.com/MurphyMc/pox> fork, which is where my 
pre-upstream work is being collected.

-- Murphy

Reply via email to