LARTC - Nov 2007 - Changing TBF rate on the fly results in drops?

Hi everybody, it is my first message here. I tried to browse  
throughout older posts but I found no immediate answer.

I need to change rate of a tbf queue while ip packets are already  
enqueued by it. If I go:

tc qdisc replace|change|link dev eth0 root tbf rate 180kbit burst 1540  
limit 15400

as results all enqueued packets seems to disappear. I can see that from a:

tc -s qdisc ls dev eth0

where backlog suddenly goes down to zero packets. Iperf performances  
also confirm that. Using custom kernel 2.6.20.1 on Ubuntu 6.06.

Where can I look to understand what is going on? Can I do something  
else to get the same result (shaping outgoing traffic dynamically and  
very often without losses)?


Thanks in advance!

____________________________________
Francesco
____________________________________

Quoting Francesco Zampognaro <zampognaro@ing.uniroma2.it>:
> Hi everybody, it is my first message here. I tried to browse throughout
> older posts but I found no immediate answer.
>
> I need to change rate of a tbf queue while ip packets are already
> enqueued by it. If I go:
>
> tc qdisc replace|change|link dev eth0 root tbf rate 180kbit burst 1540
> limit 15400
>
> as results all enqueued packets seems to disappear. I can see that from a:
>
> tc -s qdisc ls dev eth0
>
> where backlog suddenly goes down to zero packets. Iperf performances
> also confirm that. Using custom kernel 2.6.20.1 on Ubuntu 6.06.
>
> Where can I look to understand what is going on? Can I do something
> else to get the same result (shaping outgoing traffic dynamically and
> very often without losses)?
>
>
> Thanks in advance!
>
> ____________________________________
> Francesco
> ____________________________________
>
> _______________________________________________
> LARTC mailing list
> LARTC@mailman.ds9a.nl
> http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc
Hi,

I solved for now passing packets in userspace using ip_queue (iptables  
... -j QUEUE) and processing them so that the they go out respecting  
the rate which varies over time, without losses.

Still open to find a better solution in the kernel.

Regards!

____________________________________
Francesco
____________________________________