C7000 & Flex-10 - traffic between blades leaving chassis

[Alastair Jenks]

I have a c7000 chassis with flex-10 modules in Bay 1 and 2. I am using 4x 1Gb uplinks from each flex-10 module (ESX 4).

This means in the active/active config, I have 2 shared uplink sets (one per flex-10 module), each with 4 physical links. However, I have noticed that some inter-blade traffic is leaving the chassis to cross flex-10 modules.

In the instance of a file copy between blades - if the "source" and "destinations" servers happen to appear of the same flex-10 module, then no traffic leaves the chassis. However, if the "source" and "destination" servers happen to appear on different flex-10 modules then rather than the traffic crossing the flex-10 interconnect, the traffic goes out the chassis on one uplink set and re-enters the chassis on the other uplink set (wasting bandwidth back to the core switches).

Ideally, any inter-blade traffic should stay within the chassis. If I create a single SUS across BOTH flex-10 modules, then traffic IS routed across the flex-10 interconnect and NO traffic between blades leaves the enclosure. However, by creating 1 SUS across both flex-10 modules and leaving the connect type as "auto" - only 4 uplinks (from one flex-10 module) are active and the other 4 uplinks (from the other flex-10 module) are in stand-by.

How do I have all 8 uplinks active - but ensure no traffic between blades leaves the enclosure ? The situation described above seems inefficient. I am finding the flex-10 cookbook is not helping as it does not address the issue of traffic not using the flex-10 interconnect.

Has anyone else noticed this ?