tc & bandwidth limit problem

Hi everyone. I'm trying to implement the Linux 2.6 QoS capabilities to
distribute the bandwidth in my network apps, but I can't get the limits
to work properly.

For example:

tc qdisc add dev eth0 parent root handle 1: htb default 1
tc class add dev eth0 parent 1: classid 1:1 htb rate 520kbit

These sample commands should maintain output traffic to a maximum of 520
kbit/s , if I'm correct.
But the transfers here become in fact limited to approximately one
quarter (25%) of that value.
No matter what I put as 'rate', a limit seem to be enforced at 25% of
it.

(I get same results when I try cbq or tbf qdiscs, too.)

Anyone knows what the problem could be? Thanks!!
--
Pardo
Powered by: Linux, KDE, Firefox
Following: GUNDAM SEED, .hack//Roots, Kore ga Watashi no Goshujin-sama!
Playing: Silver Knights, Oblivion

pardo wrote:
> Hi everyone. I'm trying to implement the Linux 2.6 QoS capabilities to
> distribute the bandwidth in my network apps, but I can't get the limits
> to work properly.
>
> For example:
>
> tc qdisc add dev eth0 parent root handle 1: htb default 1
> tc class add dev eth0 parent 1: classid 1:1 htb rate 520kbit

Using htb default on eth can mess things up - your arp will get delayed.

Better not to use default and make a catch all IP filter.

tc filter add dev eth0 parent 1:0 protocol ip prio 99 u32 match u32 0 0
flowid 1:1

Andy.

pardo wrote:
> tc qdisc add dev eth0 parent root handle 1: htb default 1
> tc class add dev eth0 parent 1: classid 1:1 htb rate 520kbit
> These sample commands should maintain output traffic to a maximum of
> 520 kbit/s , if I'm correct.
> But the transfers here become in fact limited to approximately one
> quarter (25%) of that value.
>

Solved: new kernel build replacing CONFIG_NET_SCH_CLK_CPU with
CONFIG_NET_SCH_CLK_GETTIMEOFDAY .
Maybe the cpu cycle counter doesn't work well as clock source on this
athlon-M w/powernow, which is.. frequently hopping frequencies -_-'
--
Pardo
Powered by: Linux+KDE+Firefox
Watching: GUNDAM SEED, .hack//Roots
Playing: Silver Knights, Oblivion

pardo wrote:
> pardo wrote:
>
>>tc qdisc add dev eth0 parent root handle 1: htb default 1
>>tc class add dev eth0 parent 1: classid 1:1 htb rate 520kbit
>>These sample commands should maintain output traffic to a maximum of
>>520 kbit/s , if I'm correct.
>>But the transfers here become in fact limited to approximately one
>>quarter (25%) of that value.
>>
>
>
> Solved: new kernel build replacing CONFIG_NET_SCH_CLK_CPU with
> CONFIG_NET_SCH_CLK_GETTIMEOFDAY .
> Maybe the cpu cycle counter doesn't work well as clock source on this
> athlon-M w/powernow, which is.. frequently hopping frequencies -_-'

OK - that's the default now, the point about arp still stands you don't
really want it delayed/dropped with bulk traffic.

Talking of clocks Hz 1000 makes HTB more accurate as does changing an
option called HYSTERESIS to 0 in the source - net/sched/sched_htb.c

Hz is 250 default now, I don't know whats happened to my clock drift on
my desktop, I just noticed that on 2.6.16.11 it's now 100 times worse
than my gateway - in the past it's always been about 7 times worse and
that's on a uniprocessor/single core.

Andy.