can I replace sfq with wrr
Hello,
Currently I use htb with sfq,
/sbin/tc class add dev $IFACE classid 1:0 parent 1: htb rate ${RATE0}kbit \
ceil ${CEIL0}kbit burst 6k prio 0/sbin/tc qdisc add dev $IFACE parent 1:0 sfq
but it has one big "hole" - person which uses 10 tcp connections ( with download accelerator ) gets 10 times more bandwidth than normal user with one tcp session .
I want bandwidth share based on IP adresses ( MAC adresses ) not tcp sessions .
I've noticed that wrr could fix this problem.
I wonder if new kernel and iproute2 and :
/sbin/tc qdisc add dev $IFACE parent 1:0 wrr
instead of
/sbin/tc qdisc add dev $IFACE parent 1:0 sfq
is enough to do what I need ? or I need some more complex qdisc ?
Another question is if it's possible to limit bandwith like squid with delay pools ( so files smaller than eg: 2 MB are transmitted at full speed and larger files are transmitted : 0 - 2 MB full speed and anything more over 2MB is dropper do low priority class ) ?
How can I do this with ipchains ( kernel 2.2 ) + htb and iptables (kernel 2.4 ) + htb ?
Greets,
K.
Signature · 8 lines
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
{ Krzysztof Wittek - Gliwice Poland - ICQ: 6667414 - IRCNet: Krzychu }
{ [email protected] ftp://ftp.krzychu.net http://www.krzychu.net }
{ =====> IPv4://3646973238 IPv6://3ffe:8010:28:2::1 <===== }
- - KW6109-RIPE * + P-o-w-e-r-e-d b-y P-L-D L-i-n-u-x + * KW1-6BONE - -
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/Quoted 8 lines · Krzysztof Wittek
/sbin/tcqdisc add dev $IFACE parent 1:0 wrrinstead of
/sbin/tcqdisc add dev $IFACE parent 1:0 sfqis enough to do what I need ? or I need some more complex qdisc ?
In base it's almost enough, look at wrr doc for details. You can try esfq also [less powerfull than wrr but works simmilar]
tw
--
----------------
ck.eter.tym.pl
"Never let shooling disturb Your education"
_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
Quoted 8 lines · Krzysztof Wittek
Hello,
Currently I use htb with sfq,
/sbin/tc class add dev $IFACE classid 1:0 parent 1: htb rate ${RATE0}kbit \ ceil ${CEIL0}kbit burst 6k prio 0
/sbin/tcqdisc add dev $IFACE parent 1:0 sfq
Class 1:0 is a special class in htb. All packets queued in this class will get pas thru as fast as possible. Try class 1:1 and it will work like you want :)
Quoted 7 lines
but it has one big "hole" - person which uses 10 tcp connections ( with download accelerator ) gets 10 times more bandwidth than normal user with one tcp session .
I want bandwidth share based on IP adresses ( MAC adresses ) not tcp sessions .
There is hacked sfq qdisc : EFSQ. You can configure it so it uses only the src and/or dst address to classify the packets and not the ports. I have a local copy on www.docum.org under FAQ.
Quoted 6 lines
Another question is if it's possible to limit bandwith like squid with delay pools ( so files smaller than eg: 2 MB are transmitted at full speed and larger files are transmitted : 0 - 2 MB full speed and anything more over 2MB is dropper do low priority class ) ?
How can I do this with ipchains ( kernel 2.2 ) + htb and iptables (kernel 2.4 ) + htb ?
I think there is a patch for iptables so you can mark packets based on the size of the file where they belong to. But I'm not sure. If it exist, you can redirect big files to a queue with lower bandwidth.
Stef
Signature · 8 lines
[email protected] "Using Linux as bandwidth manager" http://www.docum.org/ #lartc @ irc.oftc.net _______________________________________________ LARTC mailing list / [email protected] http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/