[Nottingham] Bandwidth shaping

[Alex Herington]

Hi all,

I've searched around on google and I can't find anyone else's stories of 
trying this, but has anyone tried using bandwidth shaping on VLAN 
interfaces? Thanks in advance if anyone can help!

I have a box with eth0 (WAN) and eth1 (LAN running tagged VLANs on a 
NetGear switch) performing NAT between the two.

Shaping traffic coming out of eth0 is fine and dandy but shaping traffic 
out of VLAN eth1.10 (for example) seems to limit to ~128kbit/s when I 
specify any bandwidth rate, measured by pulling a file over FTP from a 
box (on the WAN side that's actually a local switch) I know I can pull 
100mbit from :/ If I remove the marking iptables rule I can pull at 
100mbit again so it's definitely the Linux traffic shaping that's at 
fault. With the same rules in place I get ~50kbit/s pulling from 
ftp.mirror.ac.uk. Again once I remove the iptables mark I can pull at 
about 3500kbit/s (so it's not a lack of WAN bandwidth in the building). 
It's just odd that the transfer speed seems proportionate to the maximum 
unthrottled bandwidth I can pull from a host! Quite bizarre considering 
the exact same commands applied to eth0 work perfectly when pushing to 
the same local FTP server.

I have VLAN header reordering turned on so that DHCPd works but I'm not 
so sure this would be causing the problem. I'm clearing all iptables 
marks and qdiscs/classes.. etc before trying these tests.

Example of the commands I'm using:

# tc qdisc add dev eth1.10 root handle 1:0 htb
# tc class add dev eth1.10 parent 1:0 classid 1:1 htb rate 512kbit
# tc class add dev eth1.10 parent 1:1 classid 1:10 htb rate 512kbit ceil 
512kbit
# tc filter add dev eth1.10 protocol ip parent 1:0 prio 1 handle 3 fw 
flowid 1:10
# iptables -I FORWARD -t mangle -o eth0 --source 10.0.10.0/24 -j MARK 
--set-mark 3

If anyone could shed any light it would be greatly appreciated :)

Regards,
Alex