robert.juric wrote:
I've been doing some research regarding the Interwork Design Guide and Data Center Infrastructure Design Guide, my question is how does the idea of the 3-tiered model hold true for virtual machine hosts. We are looking at servers with 6 connected NICs, at 20 servers a rack, that's 120 connections to the access layer per rack. Is that normal? I'm also not seeing any value in an aggregation layer. We have no campus core in our datacenter, it's servers connected to switches and switches uplinked to a core switch.

Robert

For your virtual servers you can use Nexus or 6500 switches in the access-layer with 10Gbps uplinks. So there is no reason why your virtual servers will not fit into the 3 tier model.

As for whether you need an aggregation layer. It depends. If you don't have a campus connected to the core then a perfectly feasible design is to have access-layer to core switches to WAN routers. But the problem here is scalability.

Note when i say core switches here it's really a collapsed aggregation/core.

Lets say your core  switches also do firewalling and load-balancing so you use 6500 with service modules. What happens if you then run out of ports in the core switches for the access-layer. You need to add more core switches. Now you don't just want to connect up these 2 new switches to the existing switches, that's not good design and is certainly not scalable. So you need another layer. And that's where the 3 tier  core layer comes in.

You have multiple aggregation pairs which then interconnect with each other via the core switches.

So a lot depends on how big your DC is now, how big you think it's going to get. It's really a question of flexibility/scalability.

Jon