karl and paul, expostulating
Brett D. Watson
bwatson at genuity.net
Thu Feb 20 21:55:30 UTC 1997
> >Filtering packets based on source address makes Ciscos go way slow on
> >every packet. Filtering based on destination address makes Ciscos go
> >very fast on most packets and a little slower on SYN-ACKs.
> If you enable flow switching it adds little overhead to the box.
doesn't matter. in production ios, policy routing (source based
routing) is process switched. there is code in the works to make it
fast switched. but there is a bug wherein if you do policy routing,
and you enable flow or optimum switching on the interface you're
doing pr on , it disables the policy routing.
that bug may be fixed now but in any case enabling flow switching
will *not* speed up policy routing. and if you're exporting the flow
stats, you lose anywhere from 50kpps to 100Kpps of speed.
> On a 7505
> with 2 sets of full routes and another partial set of routes (and all of
> the updates associated), that pushes some pretty significant traffic, I am
> filtering approx 25M/sec of data with 25k long extended access list. The
> total CPU load on the box is approximately 35%. Oh yeah, the box is also
> the DR for area 0 of a fairly large OSPF network (approximately 3k routes).
> Before flow switching was enabled we were running at 80% or so (not for
> more than a few minutes before we enabled flow switching though).
but i'm assuming you aren't doing policy routing in this box nor
exporting flow stats?
More information about the NANOG