Skip to contentSkip to conversation

Data rate accuracy started failing.

LARTC Peteris Krumins 2 participants 2 messages Oct 29, 2003

lore
Share & tools
lore
#0 lore raw

Dear all,

I am an experienced unix administrator and a long time ago (1+ year) I made traffic-shaping system for an ISP (~400 clients).

Each ip address requires 4 classes -
2 for incoming (internet->client), one for international traffic,
the other for local-country and border-country traffic. and
2 for outgoing (client->internet), for the same purposes.

Everything worked ok until last week, when more clients were added. The data rate (rate NNkbit) is no longer stable - sometimes it exceeds the rate specified and sometimes there is starvation. And no, there is still planty of traffic left, so its not a problem of bandwidth capacity.

The technique used is htb + sfq for traffic exceeding 0.8mbit/s, for traffics up to 0.8mbit/s as it is not cpu-intensive (but depends on HZ for calculations)

Are there any limits I am not aware of? It seems that everything works ok again if I remove some addresses (single address /32 or larger networks). I am not sure which part fails, filtering or actual shaping.

And, it is vanilla Linux 2.4.20 with qos services coming with it.

Also, i never came up with the other question: there has always been problem with tbf. For example, rate 256kbit is specified but the actual rate never exceeds 230 - 250kbit, so I add 30kbit for each rate (so 256 becoms 286kbit) so client got exactly his 256kbit.

thanks,
P.Krumins

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

#1 lore raw
Quoted 35 lines · Peteris Krumins

Dear all,

I am an experienced unix administrator and a long time ago (1+ year) I made traffic-shaping system for an ISP (~400 clients).

Each ip address requires 4 classes -
2 for incoming (internet->client), one for international traffic,
the other for local-country and border-country traffic. and
2 for outgoing (client->internet), for the same purposes.

Everything worked ok until last week, when more clients were added. The data rate (rate NNkbit) is no longer stable - sometimes it exceeds the rate specified and sometimes there is starvation. And no, there is still planty of traffic left, so its not a problem of bandwidth capacity.

The technique used is htb + sfq for traffic exceeding 0.8mbit/s, for traffics up to 0.8mbit/s as it is not cpu-intensive (but depends on HZ for calculations)

Are there any limits I am not aware of? It seems that everything works ok again if I remove some addresses (single address /32 or larger networks). I am not sure which part fails, filtering or actual shaping.

And, it is vanilla Linux 2.4.20 with qos services coming with it.

Also, i never came up with the other question: there has always been problem with tbf. For example, rate 256kbit is specified but the actual rate never exceeds 230 - 250kbit, so I add 30kbit for each rate (so 256 becoms 286kbit) so client got exactly his 256kbit.

Quoted 9 lines

thanks,
P.Krumins

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

Hi!
mmm I'm new in the forum.. but, ¿can you paste your scripts? I wanna do something like your traffic-shaping system. Sorry for my English, I'm from Spain.

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