For years, I’ve heard people talk about the fact that Fibre Channel over Ethernet (FCoE) is a technology that only Cisco is pushing or that it is a push by a dying Fibre Channel (FC) industry to extend product lives for a few more years. First of all, FC is still doing well (recent earnings from Brocade, Emulex and QLogic were all positive) and any transition from FC to Ethernet will still take many years (see my 2011 storage networking predictions). This week there were significant announcements by Intel and HP that reinforce that converged networking is much more than a Cisco initiative. FCoE will still be a minor player in the overall SAN market this year, but the ecosystem and customer adoption continue to move forward.

Intel Open-FCoE

Intel’s Open-FCoE driver that has been in development for a couple of years has been released and is supported by a large group of partners. The headline on Intel’s Press Release reads: New Technology Aims to Speed Up Cloud Adoption, Lower Costs and Environmental Impact – all it was missing was a song from wil.i.am. I have great respect for Intel, but the headline doesn’t properly explain what is going on. The early coverage that I’ve read has focused on the fact that Intel is providing “Free” software in the form of the Open-FCoE driver. For those that are familiar with storage networking history, this is similar to what was done with software iSCSI stacks which Cisco helped proliferate into Windows and Linux. There are some significant differences between what happened with iSCSI and the scenario for FCoE today.

When the “free” iSCSI software was released to the market, there was a large install base of 10/100/1000 Mbps NICs that could utilize the driver. This meant that the software solution was truly a free upgrade, especially when compared with hardware based TCP/IP Offload Engine (TOE) adapters that cost over a thousand dollars. TOEs provided minor (if any) performance improvements and possibly a few less percent CPU utilization, and ultimately customers used the NIC solutions and TOEs were only found embedded in storage arrays or appliances. FCoE, on the other hand, requires 10GbE which is far from universally deployed. Today, customers are either having to choose to add FCoE to a new server or purchasing adapters in after-market, so there aren’t many places where the Open-FCoE driver can just be thrown on existing equipment for ”free”. 10GbE volumes are much lower than 1GbE, so Intel does yet appear to have a price advantage – see current pricing from Intel NICs versus currently shipping CNAs (Note: Emulex and QLogic both offer 10Gb NICs that are lower cost than the CNA prices listed below):

• Emulex OCE10102 (2nd Gen 10Gb CNA) pricing online $590-$1393 (from Google shopping)
• QLogic QLC8242 (3rd Gen 10Gb CNA) pricing online $973-$1649 (from Google shopping)
• Intel x520 (10Gb NIC) pricing online $770-$1360 (from CDW)

The price of embedded technologies will be significantly less than the external cards, but the cost differential between vendors is likely to be a tight race for the next few years. Intel is the top selling vendor for 10GbE, but there is significant competition for LOM design wins (see recent HP OEM CNA solutions below). Intel’s PR also talks about cloud and environmental impact – converged networking is a foundational technology for many cloud environments and the energy consumption of 10GbE technologies is getting better with every generation of technology, but all of these claims can also be made by the other competitors in the space.

When it comes to interoperability, Intel’s FCoE support information can be found here – it includes Windows and Linux, Cisco and Brocade switches and EMC (getting on EMC’s support matrix is a significant hurdle to clear) and NetApp arrays. That covers all of the currently shipping FCoE switches and 2 of the 3 native storage arrays vendors (Compellent is the third), but it is missing a big player for operating systems: VMware. Emulex and QLogic were relatively easy to support with VMware since the CNA delivers FC packets to the OS, just like a FC HBA. Software solutions like Open FCoE deliver Ethernet packets; today VMware’s hypervisor can not remove the header from the packet to deliver FC and the virtual switch is not lossless (see slide 17 of my 2010 EMC World presentation). VMware and Intel have been partners for a long time, so you can expect that this will be resolved some time in the future. The press release also mentioned Oracle and Dell, conspicuously absent are IBM and HP.

Emulex and QLogic both have stickiness with their stacks for enterprise applications. If the solutions are price neutral, FC customers that look at converged networking will have a preference to stay with existing drivers to avoid having to test out the entire stack. Intel’s move is a continuation of the commoditization of IT technologies and as 10GbE becomes a standard component in more servers, there will be plenty of customers who will consider the Open-FCoE drivers. As I wrote last summer: ”The likely outcome of the transition is that Broadcom and Intel will continue to dominate. Intel will expand its presence in the storage space thanks to its Open-FCoE initiative, and Emulex and QLogic will find niches where their solutions can add value.”

HP’s FCoE Story

HP has historically been the largest sellers of FC technology, so FCoE can be seen as an opportunity to expand the storage networking market or a threat to existing revenue streams. HP participated in FCoE standards and ships a number of FCoE solutions, but sometimes delivers marketing that would push people away from FCoE. This week, HP made two announcements that show commitment to converged networking and FCoE. The first is a design win by Emulex for converged LOM on blade servers. Emulex already had this design win with 2nd generation CNA technology and this will be updated to the newer generation when it is available. The second announcement is that HP is now OEMing QLogic 3rd Generation CNAs for use in HP Proliant ML and DL series rack and tower servers. HP was reselling CNAs before, but moving to an OEM model is a stronger commitment, expanding beyond embedded solutions (like Emulex) to adapter cards that can be pre-integrated with servers or sold in after-market. Rack and stack servers are higher volume than blade servers and is a big step for HP in helping customers that want to move from FC to Ethernet. It is QLogic’s first shipping OEM of the 3rd generation CNA; according to QLogic’s new CEO Simon Biddiscome on yesterday’s earnings call, all leading server and storage OEMs vendors have the cards and a number of additional announcements can be expected in the coming months. Partnerships and OEM design wins are Emulex and QLogic’s best way to way to fend off being disrupted by the larger networking suppliers. Customers that find convergence and FCoE appealing are looking for a second source to compliment Cisco; HP has the opportunity to be that supplier and bolster its overall #2 position in the networking market.

Storage Networking for Every Server

The customer requirement is not that everything be converged, but that they have simple and cost effective solutions. If every server can have not only the connectivity required when it is deployed, but have the flexibility to adapt to the changing nature of an environment – which is even more critical for virtualized environments – IT departments will have an easier job. Emulex and QLogic will look to fend off Intel and Broadcom by adding value to the CNA offerings and fighting for design wins that make it easy for customers to continue to stay with existing driver stacks. Cisco may have helped launch FCoE and it is an integral part of its UCS servers and Vblocks with EMC, but HP, IBM, Oracle and Dell are also working with customers on converged infrastructure including FCoE solutions. It’s an exciting time for the small niche of storage networking, please feel free to post any questions in the comments or send me a note on Twitter.

[Cross-posted at Wikibon Blog]