Folks,
after spending a lot of time experimenting with HTB, I found that
filters cannot be attached to HTB classes...
The class hierarchy I use is the following (for my upstream internet link):
PRIO (with priomap that sends everything by default to
class 1:3)
/ | \
/ | \
/ | \
1:1 1:2 1:3
PFIFO PFIFO HTB qdisc (default to 4:100)
|
|
4:1
HTB class (100% bandwidth)
/ \
/ \
4:100 4:200
(50% bandwidth) (50% bandwidth)
---- HTB class HTB class--------\
/ / | \ | \ \ \
/ / | \ | \ \ \
4:110 4:120 4:130 4:140 4:210 4:220 4:230 4:240
HTB HTB HTB HTB HTB HTB HTB HTB
| | | | | | | |
| | | | | | | |
SFQ SFQ SFQ SFQ SFQ SFQ SFQ SFQ
NOTES:
Class 1:1 is used for NTP
Class 1:2 is used for DNS and ICMP
Classes 4:100 and 4:200 are used to divide bandwidth between two hosts
Classes 4:110 and 4:210 are used for audio and video traffic
Classes 4:120 and 4:220 are used for interactive traffic (such as HTTP)
Classes 4:130 and 4:230 are used for bulk traffic (such as FTP downloads)
Classes 4:140 and 4:240 are used for idle traffic (such as POP3 and SMTP)
Related tc commands:
# Traffic from $HOST_B must be sent to 4:200
tc filter add dev eth1 parent 4: u32 match ip src $HOST_B flowid 4:200
# Everything else comes from $HOST_A and is sent to 4:100
# For each host, divide the traffic according to its nature
tc filter add dev eth1 parent 4:100 protocol ip prio 10 handle
$MARK_REALTIME fw flowid 4:110
tc filter add dev eth1 parent 4:100 protocol ip prio 20 handle
$MARK_INTERACTIVE fw flowid 4:120
tc filter add dev eth1 parent 4:100 protocol ip prio 30 handle
$MARK_BULK fw flowid 4:130
tc filter add dev eth1 parent 4:100 protocol ip prio 40 handle
$MARK_IDLE fw flowid 4:140
tc filter add dev eth1 parent 4:200 protocol ip prio 10 handle
$MARK_REALTIME fw flowid 4:210
tc filter add dev eth1 parent 4:200 protocol ip prio 20 handle
$MARK_INTERACTIVE fw flowid 4:220
tc filter add dev eth1 parent 4:200 protocol ip prio 30 handle
$MARK_REALTIME fw flowid 4:230
tc filter add dev eth1 parent 4:200 protocol ip prio 40 handle
$MARK_REALTIME fw flowid 4:240
If I print statistics, the filters attached to HTB classes never
match... :-(
Does anyone know what''s going wrong ???
Stephane
_______________________________________________
LARTC mailing list / LARTC@mailman.ds9a.nl
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
On Monday 17 March 2003 22:01, Stephane Ouellette wrote:> Folks, > > after spending a lot of time experimenting with HTB, I found that > filters cannot be attached to HTB classes... > > The class hierarchy I use is the following (for my upstream internet link): > > > PRIO (with priomap that sends everything by default to > class 1:3) > / | \ > / | \ > / | \ > 1:1 1:2 1:3 > PFIFO PFIFO HTB qdisc (default to 4:100)The default class MUST be a leaf class. A leaf class is a class with no child classes. If no filter is matched, the packet is dequeued in the default class. This means that no filter is tested so your packets ends in 4:100. The filters attached to 4:100 are never checked. Stef -- stef.coene@docum.org "Using Linux as bandwidth manager" http://www.docum.org/ #lartc @ irc.oftc.net _______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Stef Coene wrote:>On Monday 17 March 2003 22:01, Stephane Ouellette wrote: > > >>Folks, >> >> after spending a lot of time experimenting with HTB, I found that >>filters cannot be attached to HTB classes... >> >>The class hierarchy I use is the following (for my upstream internet link): >> >> >> PRIO (with priomap that sends everything by default to >>class 1:3) >> / | \ >> / | \ >> / | \ >> 1:1 1:2 1:3 >> PFIFO PFIFO HTB qdisc (default to 4:100) >> >> >The default class MUST be a leaf class. A leaf class is a class with no child >classes. If no filter is matched, the packet is dequeued in the default >class. This means that no filter is tested so your packets ends in 4:100. >The filters attached to 4:100 are never checked. > >Stef > > >Stef, If I understand correctly, all filters related to the HTB leaf classes should be attached to the HTB qdisc. It would have been nice to attach filters to parent HTB classes... Stephane. _______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
On Monday 17 March 2003 22:35, Stephane Ouellette wrote:> Stef Coene wrote: > >On Monday 17 March 2003 22:01, Stephane Ouellette wrote: > >>Folks, > >> > >> after spending a lot of time experimenting with HTB, I found that > >>filters cannot be attached to HTB classes... > >> > >>The class hierarchy I use is the following (for my upstream internet > >> link): > >> > >> > >> PRIO (with priomap that sends everything by default to > >>class 1:3) > >> / | \ > >> / | \ > >> / | \ > >> 1:1 1:2 1:3 > >> PFIFO PFIFO HTB qdisc (default to 4:100) > > > >The default class MUST be a leaf class. A leaf class is a class with no > > child classes. If no filter is matched, the packet is dequeued in the > > default class. This means that no filter is tested so your packets ends > > in 4:100. The filters attached to 4:100 are never checked. > > > >Stef > > Stef, > > If I understand correctly, all filters related to the HTB leaf > classes should be attached to the HTB qdisc.No, I was only speaking about the default class. You have a filter that put some packets in 4:200. The rest of the traffic goes to the default class 4:100. But the packets are immediatly dequeued in class 4:100 so the filters attached to 4:100 are never checked. You can simulate a default class if you add a u32 filter with src 0/0. The default class is for unclassified packets. Stef -- stef.coene@docum.org "Using Linux as bandwidth manager" http://www.docum.org/ #lartc @ irc.oftc.net _______________________________________________ LARTC mailing list / LARTC@mailman.ds9a.nl http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/