[LARTC] shaping by packet count rather than bytes ?

[LARTC] shaping by packet count rather than bytes ?

[David Boreham]

In wireless networks it can be handy to shape by packet rate
rather than bytes/s (because capacity is packet-rate-limited).

Has anyone done any work on packet-rate shaping ?

Thanks.


_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] shaping by packet count rather than bytes ?

[Jens Thiele]

On  5 Okt 2007, david_list@boreham.org wrote:

> In wireless networks it can be handy to shape by packet rate
> rather than bytes/s (because capacity is packet-rate-limited).
>
> Has anyone done any work on packet-rate shaping ?

Don't know any wireless details. But I guess in the end it is very
similar to the ATM fixed cell size.
Maybe the tcatm patch [1] might give some hints.

Greetings
Jens

[1] http://ace-host.stuart.id.au/russell/files/tc/tc-atm/
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] shaping by packet count rather than bytes ?

[Peter V. Saveliev]

<skip />
> Has anyone done any work on packet-rate shaping ?
>

iptables: limit, hashlimit, dstlimit work on pps basis.

-- 
Peter V. Saveliev
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] shaping by packet count rather than bytes ?

[David Boreham]

Peter V. Saveliev wrote:
> <skip />
>   
>> Has anyone done any work on packet-rate shaping ?
>>
>>     
>
> iptables: limit, hashlimit, dstlimit work on pps basis.
>
>   
<doh> ! yes, I'd thought about that stuff but somehow
discounted it as 'not worthy' for traffic shaping.

Thanks.





_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] shaping by packet count rather than bytes ?

[David Boreham]

David Boreham wrote:
>> iptables: limit, hashlimit, dstlimit work on pps basis.
>>   
> <doh> ! yes, I'd thought about that stuff but somehow
> discounted it as 'not worthy' for traffic shaping.
Actually, I remember now why iptables doesn't work :

All it does is drop the excess packets over the limit.

That's not what I need. Rather , I want something like a
tbf that operates on packet count rather than byte count.
Thus, packets will be send at the configured rate, and
queued up if they arrive at a rate in excess of the limit.

iptables will simply drop the excess, which I'm sure will
lead to poor throughput and of course means that there's no
queue to manage to achieve a given QoS policy.

Rewriting tbf or htb to operate on packets not
bytes _seems_ like it'd do the trick.

I'm a bit surprised that nobody has made this their masters thesis ;)


_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] shaping by packet count rather than bytes ?

[Peter V. Saveliev]

В сообщении от Saturday 06 October 2007 05:16:38 David Boreham написал(а):
> David Boreham wrote:
> >> iptables: limit, hashlimit, dstlimit work on pps basis.
> >
> > <doh> ! yes, I'd thought about that stuff but somehow
> > discounted it as 'not worthy' for traffic shaping.
>
> Actually, I remember now why iptables doesn't work :
>
> All it does is drop the excess packets over the limit.
>
<skip />

Simple packet drop works for ordinary tcp congestion algorithms as a channel 
overload, and tcp decreases speed. So works RED policing filters and so on.

Beside of this, iptables can _accept_ a packet :) You can try to mark and 
queue the excess packets.

-- 
Peter V. Saveliev
_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

Re: [LARTC] shaping by packet count rather than bytes ?

[David Boreham]

Peter V. Saveliev wrote:
> <skip />
>
> Simple packet drop works for ordinary tcp congestion algorithms as a channel 
> overload, and tcp decreases speed. So works RED policing filters and so on.
>   
Well...red isn't exactly 'simple' packet drop :)
In fact, if I could combine packet-rate-limit with red or sfq I'd be 
very happy.
> Beside of this, iptables can _accept_ a packet :) You can try to mark and 
> queue the excess packets.
>   
Hmm...need more coffee...



_______________________________________________
LARTC mailing list
LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc