I was working on a HP Virtual Connect FlexFabric and Cisco FEX bake off scenario and found something which is near to the end decision. I thought I would write an article on this. Here you go. For the betterment of visualization, I have created a Visio diagram and made it available here as above.
If a new 8 port FEX is used on a set of 8 servers using the two 10G VIC, a single enclosure of 8 servers would be connected to the Fabric Interconnect using 16 cables for maximum bandwidth. That is 2 cables per server – where is the convergence and cable reduction? Is the new 8 port FEX only a reasonable choice for the new 1280 VIC and is this really overkill for most customers?
Each Cisco UCS 2208XP has 32, 10Gb ports connected through the mid plane to each half-width slot in the chasis. Therefore each server can employ a 1280 VIC with 40Gbps to each fabric in an active/active format.
This results in a total (theoretical) 640Gbps total server bandwidth constrained by a maximum 160 Gbps for a 4:1 oversubscription.
Cisco flaunts that the new VIC can deliver 80Gbps but this may be more I/O than in reality they are able to deliver due to data rate constraints on the data bus.