A client of mine has the following core network design:
1. A routed server farm access layer with L3 uplinks to dual core switches.
2. No server-farm distribution switches.
3. Closet switches are L2-trunked up to a "campus layer" (building distribution layer).
4. Building distibution layer is L3 uplinked to the dual core.
5. Enterprise edge devices are all directly connected to the core.
To summarize, the server farm switch block is routed at the access layer and does not use any distros, just L3 uplinks directly to the core. The closet switches follow the more classic Cisco design model, with L2 trunks up to a closet distro layer which is then L3 uplinked to a dual core. The enterprise edge goes directly into the core switches.
OK, now the client has purchased a new design in which a server farm distro layer will be added in between the core and the server farm access layer, and that routed access layer becomes a switched access layer instead -- just like the closet/building switch block.
The 2 core switches, which are hybrid 6500s, will be replaced by 2 native IOS 6500s with 10 Gig downlinks to the new server farm distros and 2-G Fast EtherChannel downlinks to the building distro (campus).
As I mentioned before, the enterprise edge devices -- WAN routers, VPN appliances, firewalls, etc -- are all directly connected to the existing core switches and the new design calls for more of the same. Basically, the 2 new core switches are supposed to simply replace the existing ones and wont really offer any value-added services, except for 10G downlinks.
Im wondering if the existing core switches should just be kept in place to create a separate enterprise edge switch block and just connect those directly to the core, just as the server farm and closet switch blocks will be. Right now, the 2 core switches that have all the enterprise edge devices directly plugged into them are performing some IP and IPX packet filtering. To have a true, high-speed, "packet-pushing" core, shouldnt the new core switches not have that responsibility and just act as a central point/hub for L3 forwarding?
