Fibre Channel over Ethernet (FCoE) is being heralded as the next-generation storage connectivity. Originally proposed to the standards committee ANSI T11 in April 2007 by storage vendors including Brocade Communications Systems, Cisco Systems, EMC, Emulex, IBM, Intel and QLogic, the FCoE standard is being pushed as the method to deliver a single converged network infrastructure.
With a converged FCoE network, servers use the same interface for the storage area network (SAN) and local area network (LAN). A single converged ethernet infrastructure can carry all traditional IP traffic, plus fibre channel SAN traffic and potentially all high-performance computing (HPC) traffic.
The converged network would use one set of cables and switches and consume less power, take up less physical space and be easier to manage. The datacentre manager would have a single set of hardware and software to manage and potentially there would no longer be a requirement for two teams with different skills, i.e. one team managing storage switches and fibre channel and one managing traditional LANs. This would represent significant savings on people and resources.
What is FCoE?
As the acronym implies, FCoE carries standard fibre channel traffic over an ethernet transport. Initially this will be running at 10Gbit/sec. The reality is not that simple, however, because fibre channel demands a highly reliable transport, which is not available on conventional ethernet infrastructure.
A number of enhancements to ethernet standards have been proposed under the umbrella of converged enhanced ethernet (CEE) to deliver the reliability required for FCoE. Many of these enhancements are still “work in progress” with the standards bodies. A CEE infrastructure will be mandatory for FCoE deployment which will require investment in a new switching infrastructure.
The death knell for iSCSI?
So does FCoE sound the death knell for iSCSI? Perhaps, as positioning papers from the vendors claim that FCoE will perform better than iSCSI while requiring less CPU overhead.
There is overlap between iSCSI and FCoE in that both protocols deliver block level SCSI transport over an ethernet network. However, they do address different markets. Initially heralded as an alternative to fibre channel, iSCSI has gained acceptance, primarily in cost-conscious small and medium businesses, where high performance is not required.
Unlike FCoE, iSCSI will run over a conventional gigabit ethernet infrastructure. The low cost of iSCSI is largely derived from the fact that the majority of the protocol stack runs on software. This software stack consumes significant host CPU. This CPU requirement increases significantly as the data rate increases, which will present a challenge when moving to 10Gbit networks.
It is unlikely that FCoE will displace iSCSI at the low end due to its high cost. iSCSI technology may move up from 1Gbit to 10Gbit as the cost of 10Gbit conventional ethernet falls, where it could contend with FCoE.
FCoE adoption – implications
If FCoE takes off, it could spell the end of the traditional fibre channel switch, director and host bus adapter as we know it today. It will present the need for total replacement of existing equipment. This has obvious implications for users, the channel and switch vendors.
Large corporations when deploying LAN technology may well choose FCoE-capable infrastructure. However, many are conservative with regard to change in datacentre technology, particularly where they have already invested and have something that works. Industry pundits predict that fibre channel will remain popular within the enterprise datacentre.
Today, a number of vendors have first-generation products available. These are being used for proof of concept and technology demonstrations. None of the vendors are expecting this early product to be deployed in a production environment.
According to industry analysts, a 2009 timeframe will be the earliest that anyone will seriously start looking at adoption. By this time standards will have been finalised and more “production ready” products will be available. First adoption is likely to be in the form of “top of rack” switches, where FCoE connections will be within a single rack, running from the server to the FCoE switch. Connections from the FCoE switch will use conventional SAN and LAN protocols.
QLogic forecasts that 10% of SAN ports will be FCoE by 2010, and that all Global 2000 companies will be using FCoE by 2012.
So what should datacentre managers do next? Many will welcome the adoption of FCoE if it promises to reduce cabling and connections. With large datacentres sometimes having several hundred miles of data cables, any reduction in heat, space, cabling and power will make a significant difference to infrastructure costs and management.
One recommendation would be to watch the storage hardware vendors closely as they will certainly provide indicators of how the market is moving, as will the participation of the server and application vendors. Any new standard will truly become widely adopted when the operating system and application vendors also support it.
FCoE could well be the way for organisations to future-proof both storage and datacentre requirements, but only time will tell the true value of this next-generation storage connectivity.
Stuart Bridger is manager, services and suppoer (EMEA) storage and SAN at Avnet.