[LARTC] strange behaviour of qos

[Ratel <ratel@post.pl>]

Hi

I have the following problem:
I've created qos script which shapes traffic
on outgoing interface eth1. More - less it looks like this :
------------------------CUT------------------------------------------------------ 

 #root qdisc and class for eth1
  $tc qdisc add dev eth1 root handle 1:0 htb default 19
  $tc class add dev eth1 parent 1:0 classid 1:1 htp ratel ${CEIL_UP}kbit 
ceil ${CEIL_UP}kbit

 #classes, qdiscs and filters for services
  $tc class add dev eth1 parent 1:1 classid 1:11 htb rate 90kbit ceil 
150kbit prio 0
  $tc class add dev eth1 parent 1:1 classid 1:12 htb rate 100kbit ceil 
250kbit prio 0
  $tc class add dev eth1 parent 1:1 classid 1:13 htb rate 90kbit ceil 
1250kbit prio 2


  $tc qdisc add dev eth1 parent 1:11 handle 111: sfq perturb 10    #
  $tc qdisc add dev eth1 parent 1:11 handle 112: sfq perturb 10    # 
Typical
  $tc qdisc add dev eth1 parent 1:11 handle 113: sfq perturb 10    #


  $tc filter add dev eth1 parent 1:0 protocol ip prio 0 handle 1 fw 
classid 1:11
  $tc filter add dev eth1 parent 1:0 protocol ip prio 0 handle 1 fw 
classid 1:12
  $tc filter add dev eth1 parent 1:0 protocol ip prio 2 handle 1 fw 
classid 1:13

the same thing goes for imq

 #root qdisc and class for imq0
  $tc qdisc add dev imq0 root handle 2:0 htb default 29
  $tc class add dev imq0 parent 2:0 classid 2:1 htp ratel ${CEIL_DN}kbit 
ceil ${CEIL_DN}kbit

 #classes, qdiscs and filters for services
  $tc class add dev imq0 parent 2:1 classid 2:21 htb rate 90kbit ceil 
150kbit prio 0
  $tc class add dev imq0 parent 2:1 classid 2:22 htb rate 100kbit ceil 
250kbit prio 0
  $tc class add dev imq0 parent 2:1 classid 2:23 htb rate 90kbit ceil 
1250kbit prio 2

  $tc qdisc add dev imq0 parent 2:11 handle 211: sfq perturb 10    #
  $tc qdisc add dev imq0 parent 2:11 handle 212: sfq perturb 10    #
  $tc qdisc add dev imq0 parent 2:11 handle 213: sfq perturb 10    #

  $tc filter add dev imq0 parent 2:0 protocol ip prio 0 handle 0xb fw 
classid 2:21
  $tc filter add dev imq0 parent 2:0 protocol ip prio 0 handle 0xc fw 
classid 2:22
  $tc filter add dev imq0 parent 2:0 protocol ip prio 2 handle 0xd fw 
classid 2:23

-----------------------CUT-------------------------------------------------------------------------------- 


There are more of these classes  - up to 19 (or 29 on imq0).
When I stat classes and qdiscs everything looks fine : traffic goes 
smoothly
through every class. Class 1:11 , and 2:21 are for icmp packets only.
The problem is - when I try to download some large file using http which 
goes
through 1:13 and 2:23 classes pings rise to very high values (~350 - 600 
, while normally it should be something ~5-25 ms).
The situation is getting much worse when I allow p2p traffic  (1:15, 
2:25) to pass through. Although schedulers
seem to work ,because I can browse web pages, the whole interactivity is 
lost and output (and input)
bandwidth is consumed almost totally.

my system is : 2.4.29-ow1 , additional schedulers : esfq and wrr
p2p packets are `intercepted' by p2p and ipp2p modules.
Other packets are marked this way :
--------------CUT--------------------
$IPTABLES -A PREROUTING -t mangle -i eth1 -j IMQ --todev 0
[...]
$IPTABLES -A PREROUTING -t mangle -i eth1 -p icmp -j MARK --set-mark 0xb
$IPTABLES -A POSTROUTING -t mangle -o eth1 -p icmp -j MARK --set-mark 0x1
[...]
$IPTABLES -A PREROUTING -t mangle -i eth1 -m multiport --sport 80,443 -j 
MARK --set-mark 0xd
$IPTABLES -A POSTROUTING -t mangle -o eth1 -m multiport --dport 80,443 
-j MARK --set-mark 0x3
[....]
-------------CUT----------------------------

any idea what might be wrong ?


thanks in advance
Wlodek



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