Re: [nsp] WCCP question

From: Kazu Kimura (kimura@ctc.ad.jp)
Date: Tue Dec 05 2000 - 23:37:44 EST


Lincoln,

Thank you for the information and I think we will contact Cisco distributor when we will have the plan.

Regards,
Kazu

On Tue, 05 Dec 2000 19:03:08 -0800
Lincoln Dale <ltd@cisco.com> wrote:

> Hi,
>
> At 10:52 AM 6/12/2000 +0900, Kazu Kimura wrote:
> >Hello all,
> >
> >Reviewing Cisco's document in the web page concerning WCCP, I found it
> >is a very convenient tool to save the bandwidth to upstream.
> >
> >But when we consider the introduction, there is a concern to the influence
> >to CPU utilization except the cost.
>
> characterizing the impact on cpu utilization when enabling WCCP isn't an
> easy thing to do.
> for one thing, deploying caching may result in "less packets to the
> internet" (assuming the cache is _working_ :-) ), but for another it may
> mean more packets _thru_ the router depending on where the caches are
> placed relative to the router and where the http requests are coming from
> (the web-browser).
>
> WCCP itself is implemented in most switching-paths used on cisco
> routers. it exists in dCEF on 75xx, CEF on other platforms on most IOS
> 12.xx releases and is also implemented in fast- and process-switching paths.
> it will actually operate at whatever switching-level packets are already
> being "switched at" - and won't cause the switching level to drop to a
> lower level on its own.
>
> bearing these things in mind, here's a few general statements:
>
> - use input-based WCCP over output-based WCCP where you can.
> if you think thru the logic of "how a router switches a packet", it quickly
> becomes apparent that this will mean that the router only has to switch
> intercepted packets once (as opposed to twice on output-based WCCP)
>
> - place caches into the network whereby they can "route" back to the
> web-browser
> NOT thru the router performing interception.
>
> - since a single WCCP service-group can operate on many devices at the
> same time, it is better to perform the "intercept/redirect" function
> across a bunch
> of (say) access-routers than to deploy it on a pair of core routers.
> ie. distribute the functionality across multiple devices.
>
> - for ultimate scalability, think about using switching platforms which
> support
> WCCP and perform packet-switching in hardware.
> right now, this means Catalyst 6x00.
> with the Catalyst 6500 + PFC2 + MSFC2, there is nothing stopping you from
> true wire-rate operation.
>
>
> cheers,
>
> lincoln.



This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:12:22 EDT