Re: PF (CBQ) restriction ?

["Hsiao-lung Chang" <hsiaolungchang@cksf.org>]

Scanning that at least 66,666 times per sec with max packet size on
100mbit.
With 8000 queues, scanning1.5 MB 66,666 times per sec.  One hefty
machine.
Isn't it getting a bit close to the bus speed with the overhead now.
The bus speed is a big limiting factor on the routers, and why they
don't have faster processors.
You ever time how long it took a packet to pass the interface with 50
queues vs. 8000 queues?

I would rather let them drop and be retransmitted, the overhead is too
much, and actually slows down packet delivery.
Unless its drop sensitive like switching info.

I would adjust the token regulator size and see if that increases
throughput.

Hsiao-lung

Give the gift of knowledge!
http://www.cksf.org/

-----Original Message-----
From: Henning Brauer [mailto:lists-openbsd@bsws.de] 
Sent: Friday, February 13, 2004 6:09 PM
To: misc@openbsd.org
Subject: Re: PF (CBQ) restriction ?


* evgeniy@riscom.net <evgeniy@riscom.net> [2004-02-13 23:30]:
> Henning Brauer wrote:
> >* evgeniy@riscom.net <evgeniy@riscom.net> [2004-02-13 22:10]:
> >>Henning Brauer wrote:
> >>>because there is a hard limit of 256 CBQ quques currently?
> >>i succesfull run cbq(red) with 1024 queues (i think pf can more)
> >because there is a hard limit of 256 queues currently, damnit.
> sorry, but i know about this hard limit in the kernel.
> i find link about this (in your mailling list),
> and i  tried to increase this limit and is works fine with (1024 
> queues
> maybe is can more).
> In what actually a problem in that that the limit is  CBQ_MAX_CLASSES
256
> or that itself pf-altq may not any more?

the queues are in a simple linked list. scanning it gets pretty 
inefficent with lotsoflotsoflotsof queues.

-- 
http://2suck.net/hhwl.html - http://www.bsws.de/
Unix is very simple, but it takes a genius to understand the simplicity.
(Dennis Ritchie)