Open-FCoE: Will Software Initiators Win Again?

Now that second- and, for some components, third-generation Fiber Channel over Ethernet (FCoE) equipment is coming to market, Intel is once again promoting the use of Open-FCoE software initiators with its X520 10Gbps network cards as an alternative to Converged Network Adapters (CNAs) that process the FCoE protocol traffic on-card. The question for Open-FCoE is: Will the FCoE market develop like the Fibre Channel market or the iSCSI market did?

Intel certainly hopes that FCoE will develop along the lines of the iSCSI market: Users quickly discovered that, despite the protestations of old-line SAN vendors and steely-eyed storage guys that hardware host bus adapters (HBAs) and/or TCP Offload Engine (TOE) cards would be required to get reasonable iSCSI performance, software initiators worked just fine

In fact, the rapid increases in Xeon processing power shifted the balance to software initiators. By the time we did a competitive review of iSCSI HBAs and TOE cards in 2004, Microsoft's software initiator delivered better performance than some of the HBAs. While I've long been an advocate of software iSCSI initiators, I'm not convinced that FCoE is a replay of iSCSI.

Of course, one of the biggest differences between the FCoE and iSCSI markets is that FCoE is coming into a market that already has iSCSI as an alternative. When iSCSI started taking off, it was not only a lower-cost alternative to Fiber Channel but also (and at least as importantly) ran on the same Ethernet/IP infrastructure that organizations large and small were already supporting.

FCoE, on the other hand, is a way to maintain the investment in Fibre Channel knowledge, equipment and management while taking advantage of the Ethernet juggernaut. Organizations paying thousands of dollars per switch to turn on the FCoE functions in their top-of-rack switches aren't going to be selecting Open-FCoE over a Brocade, Qlogic or Emulex CNA to save a few hundred dollars. They're going to want the CNA that's supported by SANscreen, Storage Essentials or Command Central. Storage guys manage HBAs, while network guys usually start management at the switch port.Even beyond management, the real problem with Open-FCoE is that it's only available for Windows and Linux. While Microsoft's release of a good iSCSI initator, complete with multipath support, was key to iSCSI's success, the x64 server market isn't dominated by Windows and Linux anymore. Until and unless VMware writes an FCoE initiator, organizations will have to either standardize on a CNA or use different solutions for their VMware and other servers. Given the fact that Fibre Channel users are terribly loyal to their HBA vendors, and FC users are the target market for FCoE, I don't see that as a common solution.

In the VMware environment, we see another critical difference between iSCSI and FCoE initiators: Users commonly run the Microsoft iSCSI initiator on virtual servers connecting through the VMware vSwitch rather than using Virtual Machine File System (VMFS) and the VMware initiator. This lets them convert physical servers to virtual servers without having to move the data or reconfigure the storage. The vSwitch isn't a lossless Data Center Bridging (DCB) switch, so running Open-FCoE inside a virtual server means running FCoE--which has no error recovery mechanism to deal with lost packets--is a really bad idea over a lossy network.

Disclosure: Emulex occasionally pays me to write white papers, but I don't think that's influenced my opinion here.