As per
http://www.cisco.com/univercd/cc/td/doc/product/software/ios121/121cgcr/
switch_c/xcprt2/xcdcefc.htm#xtocid278863
Configuring Load Balancing for CEF
CEF load balancing is based on a combination of source and destination
packet information; it allows you to optimize resources by distributing
traffic over multiple paths for transferring data to a destination. You
can configure load balancing on a per-destination or per-packet basis.
Load-balancing decisions are made on the outbound interface. When you
configure load balancing, configure it on outbound interfaces.
It explains Gert's point.
regards,
--zoltan
-----Eredeti üzenet-----
Feladó: Gert Doering [mailto:gert@greenie.muc.de]
Küldve: 2001. szeptember 29. 18:52
Címzett: Scott Whyte; Gert Doering
Másolatot kap: Loureiro, Rodrigo - (Bra); 'Stephen Sprunk';
cisco-nsp@puck.nether.net
Tárgy: Re: [nsp] RES: [nsp] RES: [nsp] Load balancing
Hi,
On Sat, Sep 29, 2001 at 09:40:38AM -0700, Scott Whyte wrote:
> > [..]
> > > > This must be a particular GSR weirdness (what if I want
per-packet on the
[..]
> > Sounds like a massive design problem to me. If an option is
affecting
> > the way packets are queued *outbound* it should be set on the
*outbound*
> > interface (only!) and should not have side-effects on other
interfaces.
>
> CEF is a switching/forwarding mechanism, as such it has nothing to do
with
> queuing/buffering. Not sure what your point is.
Sorry. I didn't mean "queueing", kind of a language issue. I meant
"the
way packets are distributed upon different output interfaces should not
depend on the configuration of the inbound interface, but the routing
table only - and that may be influenced by setting flags (like load
sharing algorithms, route caching, and so on)".
The fact that the configuration of an inbound interface affects all
potential outbound interfaces concerning load sharing algorithm is what
I
find problematic.
I hope I made my point more clear this time :-)
[..]
> > To rephrase: for non-GSR platforms, per-packet load-balancing works
as
> > documented, without any unobvius dependencies on inbound interface
> > settings (as long as CEF is enabled everywhere).
> You caveat kind of puts us in agreement, no?
Hmmm, I'm not sure. The way I understood you, it's important to have
"load-share per-packet" on the *inbound* interface, for *all* platforms.
What I wanted to say was "if that is so on the GSR, I don't like it.
But
for the platforms I have tested per-packet on, it is NOT necessary to
set it on the inbound interface."
The basic premise is that CEF has to be enabled everywhere anyway,
otherwise per-packet won't work.
gert
-- USENET is *not* the non-clickable part of WWW! //www.muc.de/~gert/ Gert Doering - Munich, Germany gert@greenie.muc.de fax: +49-89-35655025 gert.doering@physik.tu-muenchen.de
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:50 EDT