6500 series are top of the bill in every aspect with the 4500 only slightly trailing. Selecting the correct device for an application includes skipping functionalities that you may never need. The sole task of a core is to switch traffic, as fast as possible. Not too many features needed there This implies that a switching backbone only should be a 6500 when a 4500 cannot handle the traffic. There are reasons to select a 6500 still, growth expectations for example. You can always get the highest performance from the 6500 but then I ask you: Do you really need that huge performance or are we overdimensioning?
A similar approach can be taken for server switches. Many server farms consist of a few mission-critical servers and many others that are less vital. Personally I think that it is financial suicide to make your server-core uplink 100% non-congesting. One should rather attach only the critial stuff congestion-free and apply a sensible overbooking for the rest. Are there times when all servers on your network are fully utilized? Probably not.
One very big advantage of modular switches like 4500/6500 is that you can be very flexible in setting up the overbooking by assigning (fiber)links to either core or devices. A fixed configuration switch does not offer the same flexibility. (I would never advise copper uplinks to the core in any serious enterprise backbone.)
Personally, I would prefer a 4500 as server switch because it offers pretty costeffective Gb ports.
One could imagine though, that a single 3750 would fit in well to connect the bits and pieces that are there in every datacenter, at least for testing- or temporary servers.
Bottom of the line is that you could give the matter some thought and re-arrange your server farm in more or less the way described here. If you feel the above does not completely apply to your network, let us know.
Regards,
Leo
