Datacenters that have multiple rows of server and storage racks are best served by having multiple logical layers of switches, with the traditional design being Core, Distribution, and Access. Servers, storage, and virtualization systems work best when the systems are all in the same Layer 2 organization. The problem is that Layer 2 loop control mechanisms have issues. For example, the most consistently reliable protocol, Spanning Tree Protocol, prevents traffic from traversing half the uplinks by design. That means in a traditional switching environment, uplinks from access to distribution layer switches do not have enough bandwidth, slowing the information transfer where it is needed the most. When possible, Port Channels are used to provide multiple parallel uplinks, but this only works to a single distribution switch, and does not scale well to larger environments.
The Cisco Nexus architecture addresses this issue head on with the adoption of the Virtual Port Channel protocol. It is a special communication between redundant Distribution layer switches that allows for two switches to negotiate Port Channels with any type of Access layer switches. For example, if the Distribution layer was composed of two Nexus 5000 series switches, and the Access layer used existing Cisco Catalyst switches, you could set up Port Channels on each of the access switches, uplink them to both Nexus 5000 switches, and have all uplinks active all the time! Even better, with a setup of dual Nexus 7000 switches at the distribution layer, Nexus 5000’s and 2000’s at the access layers, all of the 10G links will be active with no loops or blocking.
