Re: [gentoo-user] traffic shaping and p2p

[Matthias Langer <mlangc@gmx.at>]

On Fri, 2005-12-16 at 13:08 +0000, Stroller wrote:
> On Dec 15, 2005, at 5:05 pm, Matthias Langer wrote:
> >
> > Well, i use azureus - and of course i know that upload-speed can be
> > limited - which is maybe in fact the best solution to my problem.
> > ...  for p2p apps - give them
> > as much bandwidth they can reasonably get but don't let them slow down
> > firefox, ssh etc. Because i want this setup just for my homenetwork, it
> > would perfectly suffice if packages get their priorities by examining
> > port-numbers. And because i want to at least partially understand what
> > i'm doing i would prefer a simple and clean setup.
> 
> I haven't used it yet, but my understanding of traffic-shaping is that 
> it's exactly what you want. I believe that other quality-of-service 
> mechanisms may require applications to be QoS aware (setting a QoS bit 
> in the packet header).
> 
> You're absolutely right in that reducing the bandwidth of the p2p app 
> isn't the ideal way to achieve what you want - I find latency in 
> browsing & surfing with BitTorrent consuming only 60% - 70% of my 
> upload - it doesn't help that other peers are continually making 
> requests of you. If you lower the bandwidth consumption in Azureous 
> then you have to remember to up it again when you go to bed - traffic 
> shaping WILL allow you to permanently maximise your p2p bandwidth, with 
> the ROUTER reducing it only when your priority services send packets.
> 
> > I know that in
> > principle the neccessairy steps to do what i wannt can be found in the
> > 'Packet Shaping HOWTO'. .... By the way,
> > there are many different packet shedulers in the kernel - and the HOWTO
> > only explains the HTP-scheduler. What about the other schedulers - can
> > they be usefull for my purposes too - and if yes, how can they be
> > configured and used ?
> 
> No idea. I hope you'll give us feedback when you've discovered more.

Ok i found out that in fact the HFSC scheduler should be the one which
does exactly what i like because it handles bandwidth and latency
seperatley. Here  is my current setup, which seems not to be ideal - ssh
is still slow when my upload is high:

# create the following tree
#                   1:
#                  1:1
#       1:10       1:20         1:30    1:40
# where 1:10 is for ssh, 1:20 for svn, 1:30 for surfing and 1:40 for
unmatched traffic

# creates the root qdisc
tc qdisc add dev eth0 root handle 1: hfsc default 40
# node 1:1
tc class add dev eth0 parent 1: classid 1:1 hfsc sc rate 441kbit ul rate
441kbit
# node 1:10 (ssh) - guaranty 1500b in 20ms with an overarall rate of
88kbit
tc class add dev eth0 parent 1:1 classid 1:10 hfsc sc umax 800b dmax
20ms rate 88kbit
# node 1:20 (svn) - guaranty 1500b in 30ms with an overall rate of of
147kbit
tc class add dev eth0 parent 1:1 classid 1:20 hfsc sc umax 800b dmax
30ms rate 147kbit
# node 1:30 (firefox) - garanty 20000b in 100ms with an overall rate of
120kbit
tc class add dev eth0 parent 1:1 classid 1:30 hfsc sc umax 20000b dmax
100ms rate 120kbit
#node 1:40  (unmatched)
tc class add dev eth0 parent 1:1 classid 1:40 hfsc sc rate 96kbit

# now that we have our qdiscs we need filters for them
# ssh
tc filter add dev eth0 protocol ip parent 1: prio 1 u32 match ip dport
22 0xffff flowid 1:10
# svn
tc filter add dev eth0 protocol ip parent 1: prio 2 u32 match ip dport
3690 0xffff flowid 1:20
# firefox
tc filter add dev eth0 protocol ip parent 1: prio 3 u32 match ip dport
80 0xffff flowid 1:30

Note that i use the u32 filter (must be enabled in the kernel) and not
iptables.

By the way, there is a very interesting article about traffic control
with qdiscs and different schedulers, in particular HFSC, written by the
author of HFSC himself in the german 'Linux Magazin' 02/2005.

I'll tell you about further experiences with HFSC - comments and
suggestions are welcome.

Matthias


-- 
gentoo-user@gentoo.org mailing list