bytelimit match (traffic shaping)

[skaya@enix.org]

>> I would really appreciate some feedback ; we use this
>> bytelimit match here on a router to control the bandwidth
>> of a high traffic web server, and it runs fine 

> well, I think it's a neat idea, but I thought that was already in the kernel
> for some time? QoS and TOS combined with the tc and ip tools from the
> iproute2 package allow you to manage different 'bandwidths' for different
> applications.

sure. but the tc+ip method requires to define a queuing discipline,
a filter to match the traffic, and a tree of classes ; altough there
are some documentations about the way it works, I found it way
overkill for most applications - and good documentation explaining
precisely the meaning of all parameters is hard to find (read : 
sometimes yet-to-be-written). most of the time, I found myself
tagging the traffic with iptables, and then using tc+ip ; with my
bytefilter match, I just have to tag the traffic, and can get rid
of tc+ip. 

CBQ and the rest of the QoS support is important, but when all you
want to do is a plain and simple bandwidth limit, it's far easier
to use bytelimit match.

in fact (to stress it again!), I developped this thing because I
could not afford to maintain a QoS based setup (developping a 
web-interface to setup bandwidth allocation and stuff seemed to
me quite complicated, whereas we already have a tool to manipulate
iptables rules in a simple way).

as I (and the people I work with) mainly use debian, I think I
will craft a kernel-patch package for debian ; I will certainly
come back with a few questions regarding the compability of 
iptables plugins :-)


regards,

Jerome Petazzoni <skaya@enix.org>