Jason R Thorpe <thorpej@zembu.com> Wrote:

>Right, so your system is spending a LOT of time fowarding packets
>(30k/sec is a pretty decent clip, after all), and since the whole
>point of fastforward is to do it immediately when the packet arrives
>(i.e. still at interrupt level), which means less time for user
>processes to run.

    The system sometimes peaks to over 40-50k :-)

>Do you see any degredation in the rate at which packets are fowarded?

    I cant tell, since the system becomes "non-responsive". The only
way for me to get back in is to reboot the system.. or wait until
the bgp sessions time out and the traffic stops.. which gives me
a few seconds to turn off fastforward.

>The reason you're losing when you crank IPFLOW_MAX up to 8192 is
>probably because you're running out of KVA space.  Can you send me
>the exact error message that occurs?  (I'm assuming it panics.)

    There are no error messages from the kernel.  We have approx 64M
allocated for KVA.

>Also, what rev of the OS are you running?

    We're running -current and 1.4.2.  We also tried this
on FreeBSD 3.4 and 4.0 and experience similar trouble.

>>System has BGP4 running with 72,000 routes in kernel.

>In any case, impressive!

>--
>        -- Jason R. Thorpe <thorpej@zembu.com>
________________________________________________________________________
Get Your Private, Free E-mail from MSN Hotmail at http://www.hotmail.com