What If FCoE Were Free?

Intel recently announced Open FCoE, a software implementation of the Fibre Channel over Ethernet (FCoE) protocol that had only previously been available in expensive Converged Network Adapters (CNA) desinged to work on standard Ethernet hardware. Assuming that you use 10GbE LAN on Motherboard (LOM), which we discussed in our last entry, it becomes prominent in servers and you might be able to have FCoE for free.Cost savings when it comes to infrastructure consolidation is not simple math yet. You have to factor in the savings of cables, interface cards and potentially infrastructure management. While those are all compelling factors, it may not be enough for an IT Manager to switch from their existing fibre channel fabric. If FCoE were available for free as a software load then the calculation changes significantly. By doing so you would replace the cost of a pair of CNAs for a pair of onboard 10GbE LOMs which could be a savings of more than $4,000 per connected host. You also make implementation much easier as the servers don't need to be opened up. Assuming that within the year every server ships with a 10GbE LOM the move to FCoE could be much more rapid than originally thought.

The first question is what do you give up? We don't yet have enough data across a variety of platforms to know for sure with Open FCoE, but it is a safe assumption that you are going to loose some CPU performance since that CPU is now handling the work of what used to be done by a CNA. Of course the follow up question will be do you need all that CPU performance? If the CPU utilization in your physical server never goes above 50% I'm not sure that doing FCoE purely in software will matter all that much. In other words the CPU is looking for something to do anyway. What you may give up with software based FCoE is some constancy of performance. This again ties into the processor load. If you load up the processor with a lot of tasks and then also have a spike in storage traffic then something, probably storage I/O, will have to give. So if you have workloads that can randomly peak a software based implementation of FCoE may not be the best choice.

My guess, and I think so is Intel's, is that there is a lot of servers that have plenty of excess CPU utilization and the software load of FCoE will be a non issue. The impact of this could be a very cost effective way to deploy FCoE. For the first time the fibre protocol can compete with iSCSI and NAS from a cost perspective. That is a winner for storage managers with a fibre investment and may be enough to keep them from deploying iSCSI or NAS. As we discussed in our article "What is TRILL's Role in FCoE Storage Networks?" all these protocols will benefit from Ethernet Fabric and Trill. If you are already a fibre channel based data center this may make the transition to the converged, lossless Ethernet Fabric even more appealing. The other winner will be the Ethernet Fabric switch manufacturers themselves. You could say that they may loose some HBA business but it may be HBA business they never were going to get because the customer was going to switch to iSCSI or NAS. I think they would rather sell the ports than the HBAs anyway.

Maybe the bigger question is FCoE in software a replacement for iSCSI or NAS? That is something we will address in a future entry.

Track us on Twitter: http://twitter.com/storageswiss

Subscribe to our RSS feed.

George Crump is lead analyst of Storage Switzerland, an IT analyst firm focused on the storage and virtualization segments. Find Storage Switzerland's disclosure statement here.