[LAD] Strange network problem ?

[Bengt Gördén]

Den Wednesday 11 March 2009 00.40.44 skrev Fons Adriaensen:
> Hello all,
>
> Today I had a strange problem when using the
> WFS system here.
>
> The 'master' PC sends a multicast message every
> 1024 samples (21.333 ms) to all 'render' PCs.
> This has to arrive on time, and when it's late
> the renderers will mute their output and report
> the error in their status messages.
>
> Today I used some ssh -X logins from my laptop
> in the WFS room to the WFS master to run Ardour
> and some other apps for a demo. All this worked
> well all the time, as it has done before.
>
> I left everything running when going for lunch
> with our visitors, and when I returned restarted
> Ardour to listen again. I got a lots 'of 'message
> too late' errors from the rendering machines, and
> interrupted sound. Strange enough this seemed to
> be related to the _volume_ of the sounds...
>
> In other words to Ardour's level meters.
> Restarting Ardour in a new ssh login did not
> help, but running it directly on the WFS master
> solved the problem. So apparently the network
> traffic required to update Ardour's meters was
> causing the delays. And clearly the whole remote
> X session was slower than normal. CPU loads
> looked normal.
>
> Now all this should be peanuts for a Gbit
> network that has no other traffic at all,
> and it worked perfectly before. I just never
> left it running for such a long time.
>
> Anyone an idea as to what is happening here,
> and how it could be cured ?

I can't say that I can help you but 19 years as a network engineer can come in 
handy.

The most obvious is that the network is overloaded in some way. Doesn't need 
to be in terms of bandwidth. It can be things like spanning-tree going 
heywire in switches. Redirects is another thing that comes to my mind. A 
sophisticated (there are things like slow pings of IPv6 addresses that fills 
up the table in a switch) DoS might also be the case.

If we assume it's the network that causes the problem you obviously have the 
switches and the routers that can cause the problems not to say firewalls. 
What network set up (equipment and such) do you have?

Is the network protected in some way? 

Have you made some performance tests? iperf is handy.

TCP:
server:> iperf -s
client:> iperf -c server

UDP:
server:> iperf -su
client:> iperf -c server -u


/bengan