If you look at traditional networking documents they recommend separating traffic types into separate physical switches. If that's not possible then using a larger core switch with multiple VLANs can accomplish the same task with some minor differences.
I planned on building separate dvswitches for each traffic type in our new vmware cluster (management, vmotion, server lan traffic, dmz traffic, etc) and then came across articles about the beauty of Network IO Control on dvSwitches. The new plan I saw out there was one large dvSwitch with several port groups assigned to various VLANs and then using shares and limits to keep traffic hogs at bay. This seemed much simpler while allowing traffic that needs more bandwidth than 1 or 2 1-Gbps ports could provide. Things like vmotion and management traffic. Plus servers obviously.
However up to now I can only find people on either camp without any up to date documentation. We're using 8-1Gbps ports, not 10-Gbps.
Can anyone say for sure if VMware prefers one way or the other? It seems as if it's moving towards the NetworkIOC with one large dvswitch, but this is non-traditional. In the real world it often makes sense to buy switches dedicated to iSCSI for example rather than plug up ports on the core, especially if not physically near the core. But in VMware there no cabling or distance limitation. The hosts are right next to each other and everything is virtual.
Am I missing something?
Thanks!