Is FCoE the emperor's new trousers?
Everyone says it looks great. Hmm.
First off, this is not going to work with your current NICs, not even the 10Gbit/s ones. Instead it will require what's in effect a Fibre Channel HBA with an Ethernet PHY (physical interface), plus the small amount of extra logic needed to support TCP/IP traffic alongside Fibre Channel.
Initially these Converged Network Adapters (CNAs) are likely to be more expensive than HBAs, admits Gervais, though he adds that having one lot of cabling and switches instead of two should still bring savings.
What about doing the job in software, and using a standard Ethernet NIC instead?
"Some people think you can do it in software," Gervais says. "There is a lot of complexity to the Fibre Channel stack though, and it's taken a few years to mature. It's not a trivial software problem."
It's also a bit different from the iSCSI hardware debacle - where the TOE developers (who include Gervais's former employer Alacritech, by the way) were outflanked by the realisation that yes indeed, it is perfectly possible to do iSCSI in software, using the spare compute cycles of a dual-Xeon class processor. That was at Gigabit speeds though - and at 10Gbit/s, things are rather different.
So if you want to run FCoE, you're going to need a pricey CNA in each server.
You will also need FCoE-aware switches to build anything more than the most basic of SANs, because FCoE is not a routable protocol. You will need one of the bridging schemes such as FCIP or iFCP to send it over a WAN, as well.
By contrast, most modern Gig Ethernet switches and routers will handle iSCSI quite happily. On the plus side, an FCoE-enhanced switch could have Fibre Channel ports as well as Ethernet, providing the link between the back-end SAN and your distributed servers.
And this isn't going to unify your LAN and SAN management - it's aimed at larger enterprises, and they are still going to need someone with the skills and experience to look after the SAN. To be fair, iSCSI also requires the administrator to learn new stuff to do with storage management, though because it's aimed more at SMEs, it's not quite so involved as learning to run a Fibre Channel SAN.
FCoE's big chance is if it can be tied into the wider process of migrating to 10Gig Ethernet. The argument will be: If you're buying 10Gig NICs, you'll probably want intelligent ones, so why not jump straight to CNAs? And if 10Gig switches can be FCoE-enabled for not too much extra cost, well, a new switch is a new switch.
That's a fairly big 'If' though, especially when standard Ethernet and IP will probably do the job in many applications.
So why are vendors pushing FCoE?
The fact is that, while the storage vendors talk of huge growth in iSCSI sales, it's easy to grow fast when you start from nothing. They have had some success with iSCSI in midsized organisations, but the grand strategy espoused by Cisco and others of using it to extend Fibre Channel out of the data centre has failed - or at least, it's looking more like a duck than a swan.
FCoE is the latest attempt to bridge the divide. We already have InfiniBand, but that's relatively short range and has only really succeeded in niches - and it's yet another different type of wire.
So some other way was needed to meet the demands for "Ethernet everywhere" while acknowledging what many experts have maintained all along - that IP just isn't the right platform for a heavy-duty SAN.
Could Ethernet become the new standard physical layer for Fibre Channel? Perhaps. After all, it was always planned that 10Gig Ethernet and 10Gig Fibre Channel would use the same connectors and modules, so running them together on the fibre (or wire) can be seen as a logical next step.
The problem is that when the pundits talk of "Ethernet everywhere," they don't mean that. What they mean is "IP everywhere" - and that's a somewhat different kettle of fish.
That's because it's not the wire that made Ethernet ubiquitous - it if was, we'd all still be using 10Mbit/s coax copper. What swept all before it was the higher-level stuff, in particular IP, but also TCP.
So while FCoE undoubtedly has a chance of stardom, as there's lots of good reasons to make one network infrastructure do the job of two, it's not yet clear that the SAN companies really understand the LAN. Perhaps they'd do better to concentrate on fixing the management layer, so that iSCSI becomes a practical proposition for today's Fibre Channel users, as well as for SMEs.