Skip to contentSkip to conversation

Sharing incoming traffic

LARTC Julián Muñoz 4 participants 9 messages Mar 21, 2002

lore
Share & tools
lore
#0 lore raw

I've done my first test with ingress,

2 ftps, and I've seen that the bandwidth is not shared "very well".

From the point of view of a user, his transfer stops suddenly, and restarts 20 seconds (or more!) later. Then the other has to wait !! I observ a kind of feedback process, beeing the interval of stopped traffic bigger each time, during the transference.

The bandwidth is limited to 64.000 bit per second, killing packets.

In fact it is not a real ethernet link, and the filter is on a vmware machine computer, so maybe this test is not valid.

Anyone knows more about this behaviour ??

Could I optimize it playing with burst and mpu ?

Or am I doing something really bad ?

Thank you,

Here's my filter:

iptables -A PREROUTING -i eth0 -t mangle --protocol all -j MARK --set-mark 1

tc qdisc add dev eth0 handle ffff: ingress

tc filter add dev eth0 parent ffff: protocol ip prio 5 handle 1 fw \
police rate 8000bps burst 10k mpu 64b drop flowid :1

Signature · 16 lines
      __o
    _ \<_
   (_)/(_)

Saludos de Julián
EA4ACL
-.-

Foro Wireless Madrid
http://opennetworks.rg3.net



_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
#1 lore raw

For me ingress works great.
ok, i more or less copy/pasted from the "wondershaper" :)

I guess it's not good to have a rate of 8000bps AND a burst of 10k

I'm also not shure, if the iptables-marks get noticed, but it seems so, as you said there is a shaping effect. but as your iptable-rule is so generally i'd say there's no reason not to use the appropriate u32 filter instead of fw.

VMWare shouldn't be the problem, at least i never had ones.

Tell me, if tuning rate/burst helped

Greetings

Tobias

Quoted 25 lines · Julián Muñoz

I've done my first test with ingress,

2 ftps, and I've seen that the bandwidth is not shared "very well".

From the point of view of a user, his transfer stops suddenly, and restarts 20 seconds (or more!) later. Then the other has to wait !! I observ a kind of feedback process, beeing the interval of stopped traffic bigger each time, during the transference.

The bandwidth is limited to 64.000 bit per second, killing packets.

In fact it is not a real ethernet link, and the filter is on a vmware machine computer, so maybe this test is not valid.

Anyone knows more about this behaviour ??

Could I optimize it playing with burst and mpu ?

Or am I doing something really bad ?

Thank you,

Here's my filter:

Quoted 27 lines

iptables -A PREROUTING -i eth0 -t mangle --protocol all -j MARK --set-mark 1

tc qdisc add dev eth0 handle ffff: ingress

tc filter add dev eth0 parent ffff: protocol ip prio 5 handle 1 fw \
police rate 8000bps burst 10k mpu 64b drop flowid :1

Signature · 16 lines
      __o
    _ \<_
   (_)/(_)

Saludos de Julián
EA4ACL
-.-

Foro Wireless Madrid
http://opennetworks.rg3.net



_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

#2 lore raw
In reply to Julián Muñoz

the "wondershaper" by one of the authors of the LARTC-HOWTO:
http://lartc.org/HOWTO//cvs/2.4routing/output/2.4routing-15.html
at the very end of this html-page simply copy/paste the ingress-line and the next line which is the u32 filter for the ingress-qdisc. that worked perfect for me :)

Quoted 5 lines

Thank you Tobias,

Quoted 2 lines

For me ingress works great.
ok, i more or less copy/pasted from the "wondershaper" :)

Quoted 9 lines

Eh?

Quoted 1 line

I guess it's not good to have a rate of 8000bps AND a burst of 10k

I have set burst to 3k, and this has not helped, I still see the "not nice" sharing at the input.
Maybe I should increase burst ??

well, what do you want? i guess i want to "test" the ingress-qdisc, and as far i understand the whole thing it's no good idea to set a specific rate but at the same time allow a burst that's x*specific-rate. i always set my burst to something like specific-rate/10 or so in your example with 8000bps rate, i'd test a burst of 800 or even less

Quoted 6 lines
Quoted 2 lines

I'm also not shure, if the iptables-marks get noticed, but it seems so, as you said there is a shaping effect.

You say the if I use u32 maybe this effect would disappear ?

like i said: i dont' know exactly :) i'm not in this c / hardcore-kernelprogramming stuff, so i don't know for shure which code comes first in case of ingress: the qdisc-stuff or the iptables stuff. if the iptables stuff comes first (and it seems so) then everything's ok. let it be.
the "effect" has nothing to do with your filter, more with your qdisc (i guess :)

Quoted 9 lines
Quoted 3 lines

but as your iptable-rule is so
generally i'd say there's no reason not to use the appropriate u32 filter instead of fw.

The problem is that u32 is not well documented :-(
I had to install all the ipchains and after iptables, because was unable to do anything "coherent" with u32...

like mentioned above: the last line of code in the html-page is the u32-filter line you need (and which does the same as your iptables-mark-rule)

good luck

tobias

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

#4 lore raw
In reply to Julián Muñoz

Ok,

what I really wanted was to prioritize some kinds of incoming traffic (DNS or HTTP) (or penalyze some other, which is same), exactly for what you say, "gain interactivity on a heavy loaded link"

but doing test, I saw this strange "non fairness sensation", which is maybe problematic ??

That's why I ask if someone has already seen this behaviour, or it is specific of my system ? The test is very easy, do 2 ftp in parallel.

I am going to recompile 2.4 on my main computer, hopping this is a caused by vmware :-)

Quoted 14 lines · Tobias Geiger

I reread your Subject line and noticed you meant SHARING. i always read SHAPING.
Make sure u can't actually SHARE your incoming bandwith. you can just throttle it down (to gain more interactivity on a heavy loaded link)

I hope we didn't missunderstand each other

Tobias

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Signature · 14 lines
      __o
    _ \<_
   (_)/(_)

Saludos de Julián
EA4ACL
-.-

Foro Wireless Madrid
http://opennetworks.rg3.net

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
#8 lore raw
In reply to Julián Muñoz

You say that IMQ could be attached to a device to the incoming side ??

Quoted 9 lines · Martin Devera

With IMQ you can ;)

Quoted 2 lines

Make sure u can't actually SHARE your incoming bandwith. you can just throttle it down (to gain more interactivity on a heavy loaded link)

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/

Signature · 14 lines
      __o
    _ \<_
   (_)/(_)

Saludos de Julián
EA4ACL
-.-

Foro Wireless Madrid
http://opennetworks.rg3.net

_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/