Hello Dan,

I thought this (your concept) would be fine as well but the WCCP configuration guide for the Catalyst 4500 switch states that the cache engines should not be separated by a Layer 3 hop. That's the part that confuses me.

http://www.cisco.com/en/US/docs/switches/lan/catalyst4500/12.2/46sg/configuration/guide/wccp.html#wp1000978

Restrictions for WCCP

The following limitations apply to WCCP:

•A service group can comprise up to 32 content engines and 32 routers.

•All content engines in a cluster must be configured to communicate with all routers servicing the cluster.

•Up to 8 service groups are supported at the same time on the same client interface.

•The L2 rewrite forwarding method is supported, but GRE encapsulation is not.

•Direct L3 connectivity to content engines is required; L3 connectivity of one or more hops away is not supported.

Both are correct, for software based IOS routers (7206, 38xx, 28xx, 18xx) and Cat6K+SUP720, using GRE-Redirect, the WCCP client can be multiple L3 hops away.

For Hardware based redirection, including most Catalyst based platforms, you have to use L2-redirect which requires that your WCCP client be L2 adjacent which won't work in your situation unless you have L2 spanning both DCs.

Can you move WCCP to a platform that supports GRE-redirect like a 7206 or ISR based router? Another possibility would be to use a L4-7 switching plafform

Hope that helps,

Dan

Thanks for the clarification. Using a layer 7 switch (load balancer) will introduce another point of failure. Due to the fact that there is a layer 3 connection between DCs, I will not be able to cluster the LBs.....giving me another component to worry about :-)

A solution like GSS could help in this regard but if I will be introducing DNS load balancing into the mix, WCCP may not be the most attractive option.

I appreciate your help.

If you can't use a IOS based GRE-redirect platform, another other option would be to do 2 WCCP clusters locally using L2-redirect. That way you can still do your redirects, however if a proxy fails, 1/2 of your traffic doesn't get proxied unless you doubled up locally on hardware.

However, this also reduces DC to DC traffic that WCCP would increase due to the load balancing, however adds incremental up front $$ to the implementation.

Dan

Thought about that too......unfortunately, the budget does not allow me to double up on hardware.

The key here is that I also have to consider DC redundancy as the network has quite a few remote sites. So my ideal choice will achieve both hardware and site redundancy at the same time.