Skip to contentSkip to conversation

CBQ or HTP accuracy

LARTC Thierry Coutelier 2 participants 3 messages Dec 4, 2002 – Dec 5, 2002

lore
Share & tools
lore
#0 lore raw

Hello,

I am rate limiting traffic going to some IP addresses using CBQ
(and want to change to HTB sometime).

While testing with different packet sizes I found out that I could go above the
rate limit I set. (even with HTB)

The values I use are: rate: 768kbit weight 77kbit allot 10000 avpkt 1512 I did not change the cell or other parameters.
My root qdisc is set to 40MBit.

With those values I am able to go up to 840kbit/s with packets sizes of 612 or 1224. With other values it's less except for very low values.

Changing the allot value (If I understood well allot*weight=max packets per round)
it only gets worst (tested with 1000 and 50000).

AVPKT is only used for maxidle so it should not change anything ?

What values should I tweak to have the max rate never exeeded and still be able
to come very near my max-rate using 1460 bytes packets ?

Would HTB be more accruate ?

Thanks for the HELP.

Signature · 9 lines
Thierry Coutelier      /'\ ASCII Ribbon Campaign
http://www.linux.lu    \ / No HTML/RTF in email
.                       ×  No Word docs in email
.                      / \ Respect for open standards


_______________________________________________
LARTC mailing list / [email protected]
http://mailman.ds9a.nl/mailman/listinfo/lartc HOWTO: http://lartc.org/
#1 lore raw
Quoted 25 lines · Thierry Coutelier

Hello,

I am rate limiting traffic going to some IP addresses using CBQ
(and want to change to HTB sometime).

While testing with different packet sizes I found out that I could go above the
rate limit I set. (even with HTB)

The values I use are: rate: 768kbit weight 77kbit allot 10000 avpkt 1512 I did not change the cell or other parameters.
My root qdisc is set to 40MBit.

With those values I am able to go up to 840kbit/s with packets sizes of 612 or 1224.
With other values it's less except for very low values.

Changing the allot value (If I understood well allot*weight=max packets per round)
it only gets worst (tested with 1000 and 50000).

AVPKT is only used for maxidle so it should not change anything ?

Quoted 9 lines

What values should I tweak to have the max rate never exeeded and still be able
to come very near my max-rate using 1460 bytes packets ?

Would HTB be more accruate ?

Thanks for the HELP.

Thierry,

to my knowledge, HTB is more predictable than CBQ and easier to set up.

If you want more accuracy, I suggest you modify a file in the kernel
source: include/net/pkt_sched.h

Change the #define for PSCHED_CLOCK_SOURCE from PSCHED_JIFFIES to
PSCHED_CPU.

I've read this suggestion on the list but never tried it yet. Note
that this is valid only for CPUs that support a TSC (Pentium and up).

Regards,

Stephane

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

#2 lore raw
In reply to Thierry Coutelier

I already tried it and it does not improve much the behaviour.

With which parameters should I play in HTB and CBQ ?

Quoted 53 lines · Stephane Ouellette
Quoted 25 lines · Thierry Coutelier

Hello,

I am rate limiting traffic going to some IP addresses using CBQ
(and want to change to HTB sometime).

While testing with different packet sizes I found out that I could go above the
rate limit I set. (even with HTB)

The values I use are: rate: 768kbit weight 77kbit allot 10000 avpkt 1512 I did not change the cell or other parameters.
My root qdisc is set to 40MBit.

With those values I am able to go up to 840kbit/s with packets sizes of 612 or 1224.
With other values it's less except for very low values.

Changing the allot value (If I understood well allot*weight=max packets per round)
it only gets worst (tested with 1000 and 50000).

AVPKT is only used for maxidle so it should not change anything ?

Quoted 9 lines

What values should I tweak to have the max rate never exeeded and still be able
to come very near my max-rate using 1460 bytes packets ?

Would HTB be more accruate ?

Thanks for the HELP.

Thierry,

to my knowledge, HTB is more predictable than CBQ and easier to set up.

If you want more accuracy, I suggest you modify a file in the kernel
source: include/net/pkt_sched.h

Change the #define for PSCHED_CLOCK_SOURCE from PSCHED_JIFFIES to
PSCHED_CPU.

I've read this suggestion on the list but never tried it yet. Note
that this is valid only for CPUs that support a TSC (Pentium and up).

Regards,

Stephane

Signature · 9 lines
Thierry Coutelier      /'\ ASCII Ribbon Campaign
http://www.linux.lu    \ / No HTML/RTF in email
.                       ×  No Word docs in email
.                      / \ Respect for open standards


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