Ethernet Adapts for Data Center Applications

Ethernet inherently doesn't provide the ability to multi-path because STP is blocking our redundant links to mitigate loops in the network. So if you are implementing Fibre Channel over Ethernet and have promised your SAN team that the network won't lose their Fibre Channel frames, the next hurdle will be multi-pathing. (See previous post that discussed the ways Fibre Channel and Ethernet don't get along, and why Data Center Bridging is the answer.)

Single - hop as its name implies is a topology where the Fibre Channel traffic is encapsulated in Ethernet for a single hop, from the host to the switch. The switch strips the Ethernet frame off of the Fibre Channel frame and transmits it as a native Fibre Channel frame into the SAN. In this topology multi-pathing isn't a real consideration because the best practice would be to connect the FCoE enabled host to different physical switches which are in turned connected to the A fabric and B fabric in the SAN. Multi-pathing would be handled on the host and VLAN/VSAN isolation provides the separation on the Ethernet topology.

Multi-hop topology

In a multi-hop topology, the design becomes more interesting as our Fibre Channel network isn't connected to the first hop switch and we now have multiple Ethernet paths through the network. As we discussed earlier, STP will block links in our network to provide a loop free topology which won't meet Fibre Channel's needs. Enter Cisco Fabric Path and Transparent Interconnections of Lots of Links (TRILL). We'll discuss Cisco Fabric Path and TRILL in more detail in the next post, but know that these two technologies provide the ability to run Ethernet networks without STP by doing MAC based routing.

Ultimately we'll see more adoption of FCoE over time as the technologies mature, operational best practices are established and mature and those lessons learned from the school of hard knocks are identified. We observed the same trend with SNA to IP migrations (RSRB to DLSw and then DLSW+ anyone?) and VoIP. The industry as a whole has the experience and wherewithal to reap the same cost benefits and reduction of equipment in the data center as we have with the other consolidated technologies. Will it be easy, not necessarily, but as we've learned from history we've been able to avoid many of the early challenges we saw with convergence. QoS as an example is much better known and understood than it was years ago and will become a key technology in the data center.