[c-nsp] Stacking 3750X vs diverse 4948E

[David Coulson]

In a datacenter environment, we typically deploy 4948 top-of-rack 
switches with L2 uplinks to our 6500 core - Systems get connections into 
two different switches and rely on OS NIC bonding (mostly Linux) to 
support switch failures. Switches running STP and in the last four years 
we've had no issues with this design (including failures of systems 
connected to diverse switches).

A new proposed configuration utilizes stacked 3750X switches, where 
servers would be connected to multiple switches within the same stack. I 
have next to no experience in the low-end switches that do stacking, but 
from a general risk management perspective, it seems like a many eggs 
and single basket configuration.

Does anyone have any solid experience with 3750X switches, or stacking 
in a datacenter in general? I've seen plenty of stacks for 
closets/end-users, but I don't see many in a top-of-rack config. Is 
Cisco stacking typically 'reliable', in that when a switch fails it will 
leave the remainder of the stack functional? What about a software 
issue? Does the whole stack crap out and reload, or does the master just 
fail and a new one get elected?

I realize it's a pretty broad question, but it boils down to - Is a 
stacked switch config significantly less reliable/resilient/available 
than two TOR switches?

David