Tuesday, November 28, 2017

Hyper-collapsed core design?

We're designing a one-rack colo space that would basically be a data center in a box. The servers are going to be with a hyper converged cluster (apps, storage, etc all on the same platform.)

One thought I had was basically rack a pair of beefy datacenter switches in there, and collapse everything (access, core, distribution, dmz, etc) into that single layer, segment everything with vlans/private vlans, and vrf (for the external segment)

Then you'd basically have one big trunk going down to the server cluster, one big trunk going north to the firewall, and your links to external peers in their own vrf.

That way everything truly is handled by the core switches, so basically you have one cluster of servers, one pair of switches, and one pair of firewalls.

I know this may not meet any best practice design guide but it's kinda out of the box experimenting... what do you think? Even eliminates a separate box for edge routers too... most/all dc switches will do bgp peering and it won't be full tables... default only



No comments:

Post a Comment