While conceptually this is all true, the reality can be slightly
different.
If you look at how many actual cache entries a 3600 series router would
be expecting to see (given the size of a pipe that it can handle, and
the commensurate data source/sink it would be attached to) fast-caching
can, and in my experience does, perform better overall then CEF, taking
less CPU and memory.
For that matter I know of quite a few 7200 and 7500 (distributed and
not) deployments where fast/optimum-caching works quite well, and it can
be argued if CEF would really add much benefit.
RPF does offer some benefits, but again you need to look at your
applications to see if RPF (with or without the enhancements) even makes
sense.
David
-----Original Message-----
From: Rubens Kuhl Jr. [mailto:rkuhljr@uol.com.br]
Sent: Monday, April 23, 2001 1:42 PM
To: Dan Hollis
Cc: cisco-nsp@puck.nether.net
Subject: RE: [nsp] CEF
Even on processor-based routing, CEF should have a better stability to
changing traffic patterns than demand-based fast-caching. Also, CEF
pruning
instead of clearing the fast-cache should show a better response to
route
instability, taking lesse CPU cycles to recover steady state.
A possible gain is using RPF instead of access-lists when no other
filtering
besides spoofing is required.
Rubens Kuhl Jr.
-----Original Message-----
From: Dan Hollis [mailto:goemon@anime.net]
Sent: Monday, April 23, 2001 5:24 PM
To: Pegg Damon
Cc: 'Stefan Simko'; cisco-nsp@puck.nether.net; jared@puck.nether.net
Subject: RE: [nsp] CEF
On Mon, 23 Apr 2001, Pegg Damon wrote:
> In general, CEF does wonders for your cpu util but at heavy memory
cost.
I haven't noticed any improvement on 36xx'en, regardless of traffic load
or cpu load. I guess CEF only helps on other architectures, 7xxx and
above?
-Dan
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:35 EDT