Skip to contentSkip to conversation

Limited number of bands in PRIO qdisc

LARTC Patrick Reinhardt 3 participants 5 messages Aug 22, 2007 – Aug 27, 2007

lore
Share & tools
lore
#0 lore raw

Hello,

is it possible that the number of bands for the PRIO qdisc is limited to 16?

tc qdisc add dev $DEVICE root handle 1: prio bands 16 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1

succeeds but

tc qdisc add dev $DEVICE root handle 1: prio bands 17 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1

returns: 'RTNETLINK answers: Invalid argument'

Is there any possibility to raise the number of bands to a higher value?

Thank you in advance.

Patrick
_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

#1 lore raw

I answer randomly but,
perhaps you have to give 17 parameters as well for the priomap. you have only 16 of them in your second example.
Vincent.

Quoted 20 lines · Patrick Reinhardt

Hello,

is it possible that the number of bands for the PRIO qdisc is limited to 16?

tc qdisc add dev $DEVICE root handle 1: prio bands 16 priomap 1 2 2 2 1 2 0 0
1 1 1 1 1 1 1 1

succeeds but

tc qdisc add dev $DEVICE root handle 1: prio bands 17 priomap 1 2 2 2 1 2 0 0
1 1 1 1 1 1 1 1

returns: 'RTNETLINK answers: Invalid argument'

Is there any possibility to raise the number of bands to a higher value?

Quoted 7 lines

Thank you in advance.

Patrick
_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

#2 lore raw
In reply to Patrick Reinhardt
Quoted 9 lines · Patrick Reinhardt

Hello,

is it possible that the number of bands for the PRIO qdisc is limited to 16?

tc qdisc add dev $DEVICE root handle 1: prio bands 16 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1

succeeds but

Quoted 8 lines

tc qdisc add dev $DEVICE root handle 1: prio bands 17 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1

returns: 'RTNETLINK answers: Invalid argument'

Is there any possibility to raise the number of bands to a higher value?

Thank you in advance.

I got a better error message than that -

"priomap" index > TC_PRIO_MAX

Greping for TC_PRIO_MAX shows it's in both kernel and iproute2's copy of the headers - I don't know if you need to do both or not - but I would and recompile kernel and tc

/include/linux/pkt_sched.h:#define TC_PRIO_MAX 15

Andy.
_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

#3 lore raw
In reply to Patrick Reinhardt

Hi!

Am Mittwoch, 22. August 2007 21:27:52 schrieb Andy Furniss:

Quoted 9 lines
Quoted 1 line · Patrick Reinhardt

returns: 'RTNETLINK answers: Invalid argument'

I got a better error message than that -

"priomap" index > TC_PRIO_MAX

Greping for TC_PRIO_MAX shows it's in both kernel and iproute2's copy of the headers - I don't know if you need to do both or not - but I would and recompile kernel and tc

Thank you very much for this hint. Unfortunately my current idea would require about 1000 bands:

To emulate a wireless network I have to assign a packet loss depending on the target IP address. The idea was to use a PRIO qdisc with n bands for n nodes and use filters to assign packets to the appropriate band depending on the target address.

Since I am unsure about side-effects of an alteration of TC_PRIO_MAX to a signifficantly higher value, an alternative would be helpful.

Thanks in advance.

Patrick

_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc

#4 lore raw
In reply to Patrick Reinhardt
Quoted 15 lines · Patrick Reinhardt

Hi!

Am Mittwoch, 22. August 2007 21:27:52 schrieb Andy Furniss:

Quoted 9 lines
Quoted 1 line · Patrick Reinhardt

returns: 'RTNETLINK answers: Invalid argument'

I got a better error message than that -

"priomap" index > TC_PRIO_MAX

Greping for TC_PRIO_MAX shows it's in both kernel and iproute2's copy of the headers - I don't know if you need to do both or not - but I would and recompile kernel and tc

Thank you very much for this hint. Unfortunately my current idea would require about 1000 bands:

Quoted 7 lines

To emulate a wireless network I have to assign a packet loss depending on the target IP address. The idea was to use a PRIO qdisc with n bands for n nodes and use filters to assign packets to the appropriate band depending on the target address.

Since I am unsure about side-effects of an alteration of TC_PRIO_MAX to a signifficantly higher value, an alternative would be helpful.

I suppose you could try ifb - but there is probably a define somewhere needing to be changed if you want 1000, but then you could have 16 prio on each.

Andy.

_______________________________________________
LARTC mailing list
[email protected]
http://mailman.ds9a.nl/cgi-bin/mailman/listinfo/lartc