By Dave Simpson
In April, a group of vendors proposed the Fibre Channel over Ethernet (FCoE) standard, which would allow Fibre Channel traffic to flow over standard Ethernet networks. The proposal generated immediate controversy, with iSCSI proponents-and some industry analysts-arguing that it was little more than an attempt by Fibre Channel vendors to stall adoption of iSCSI (which also allows storage traffic to run over Ethernet networks). FCoE proponents counter-argued that the standard was necessary to enable a unified, or converged, Ethernet infrastructure that could handle LAN, SAN, and clustering traffic while preserving existing investments in Fibre Channel.
Before we delve into the arguments in favor of FCoE, it should be noted that a ratified standard is not expected until at least mid-2008, and that products (host bus adapters, switches, etc.) supporting the standard probably won’t be available until at least 2009.
Nevertheless, that time frame is within the planning cycles of some IT organizations that may benefit from FCoE- particularly large enterprises that have invested heavily in both Fibre Channel and Ethernet infrastructures.
“The goal of FCoE is a converged Ethernet that handles LAN, SAN, and clustering traffic in order to reduce costs and management,” says Taufik Ma, vice president of marketing in Emulex’s Intelligent Network Products group. “Unified, or converged, Ethernet has been the end goal since at least 2000, but it never happened.”
The reasons for that, according to Ma, are twofold. First, mixing storage and LAN traffic on the same fabric “gives users the jitters. Even with iSCSI, there are usually two separate fabrics-one for SAN, one for LAN traffic.” Second, large enterprises that have invested heavily in Fibre Channel are generally unwilling to rip-and- replace that infrastructure in favor of iSCSI (which is partly why iSCSI has not made many inroads at large enterprises).
“FCoE is a way to do server convergence with existing Fibre Channel SANs,” says Lisa Guess, senior product manager in Brocade’s emerging business group. “It will enable you to leverage the 10Gbps Ethernet infrastructure for clustering, storage, and network connectivity, instead of using three separate, dedicated NICs or HBAs in special-purpose networks. It boils down to not having to rip-and-replace your existing Fibre Channel infrastructure, and protecting your investments.” (Brocade is participating in the FCoE standards group.)
“FCoE fits in several scenarios that iSCSI doesn’t, and it would provide a common Ethernet infrastructure,” says Rich Ramos, chief technologist and director of systems and protocols at Xyratex, which is also participating in the FCoE standards development. “FCoE will probably provide better performance than iSCSI because iSCSI has the overhead of the TCP/IP stack,” Ramos says. “In addition, from the physical layer up FCoE still looks like Fibre Channel and SCSI, so all existing software will work in an FCoE environment.” However, performance comparisons between FCoE and iSCSI are premature at this time.
How it works
The FCoE specification defines how to encapsulate a Fibre Channel packet in an Ethernet packet in the form of a payload. An FCoE host adapter, for example, would send FCoE packets into the fabric, and the switch would treat it like any other Ethernet packet. “This enables you to attach a converged Ethernet fabric into an existing Fibre Channel SAN, so you can hold on to your existing investments,” says Emulex’s Ma.
One technical hurdle is the need to convert Ethernet from a “lossy” protocol (that drops packets, or frames, under congested network conditions) to a “lossless” protocol, which the ANSI T11 standards group that’s developing the FCoE spec is working on. This enhancement, which will be based on “pause frame”-or “pause-based flow control”-technology, would enable the level of reliability required for storage transmissions that is currently provided by the Fibre Channel protocol.
“The industry is enhancing Ethernet so you can have more than one traffic class-for example, SAN and LAN classes, where FCoE traffic has lossless behavior,” says Ma.
“TCP has proven to be cumbersome for storage, but with some technology advances you’ll be able to do lossless transmission so you can preserve your Fibre Channel investment while taking advantage of your Ethernet infrastructure for storage,” says Harry Mason, director of industry marketing at LSI, another participant in the FCoE development effort.
Of course, implementing FCoE would require changes to existing gear. For example, specialized multi-protocol network interface cards (NICs) or HBAs-or what Emulex refers to as “converged network adapters”-would be required on Ethernet-attached hosts. And Ethernet switches will have to support the FCoE protocol. In addition, an FCoE-FC gateway device may be required, although it could be integrated into Ethernet switches. (The Fibre Channel infrastructure-both hardware and software-would remain unchanged in an FCoE scenario.)
Such products aren’t expected until at least 2009, although one vendor is already shipping an FCoE product. Finisar last month released a test and analysis tool that supports the FCoE protocol, and some of the vendors working on FCoE products are already using the analyzer in their development projects. Finisar is a voting member of the T11 FCoE group. Paul Hansen, vice president of marketing at Finisar, predicts the standard will be ratified in 12 to 18 months.
Although implementing FCoE will require a lot of changes, proponents argue that this won’t be a big hurdle because migration to FCoE will occur along with the migration from 1Gbps Ethernet to 10Gbps Ethernet. As such, end users will already be upgrading their infrastructure.
“Upgrading will not be significant because FCoE support will come along with the migration to 10Gbps Ethernet. Companies will already be buying new equipment,” says Finisar’s Hansen.
For more on the FCoE spec, visit the ANSI T11 Website at www.t11.org.