This article can also be found in the Premium Editorial Download "Storage magazine: Exploring the solid-state storage advantage."
Download it now to read this article plus other related content.
Fibre Channel over Ethernet is speeding along the certification path, and now is the time to determine what it can do for you.
The Ethernet part of the protocol isn't just any Ethernet, but a special, still-to-be-ratified Data Center Ethernet (DCE). To make Ethernet suitable for Fibre Channel transport, the 802.1Q IEEE standard is being modified to accommodate data center traffic to improve its priority traffic flow and allow it to operate in a lossless manner (no dropped packets).
The goal is to deliver FC a different network protocol leveraging an Ethernet fabric, while maintaining the same or better performance that Fibre Channel-only networks have enjoyed. Based on the IEEE modifications currently being considered, our best guess puts the timeframe for the protocol's delivery around Q4 2008. The protocol must be ratified by the FC standards body and the INCITS T11 Technical Committee.
| we understand FCoE, the real question is whether or not we need it. Why not jump to iSCSI? Basically, FCoE provides an elegant way to migrate Fibre Channel to Ethernet, while protecting existing FC investments and skill sets. While iSCSI utilizes TCP layered on top of IP, neither FCoE nor iSCSI is routable. InfiniBand also looked promising in this area, but has so far only found a home in high-performance computing environments and hasn't seen widespread adoption otherwise.
So why do we want to converge Fibre Channel and Ethernet into a single fabric? By combining FC and Ethernet, a single cable and a single card can replace current network interface and host bus adapter (HBA) cards. And because every new technology needs a three-letter acronym, the resulting interface card will be known as a Converged Network Adapter (CNA). These will feed switches or directors, and accommodate FC or FCoE and Ethernet over 10Gig links. In addition, the same switches and directors will handle both protocols while simultaneously allowing storage and networking domains to control their traffic independently. This preserves the separation of the storage and networking management domains while consolidating hardware.
Who's involved in this effort? From a storage perspective, we should look at the usual suspects: Brocade, Cisco Systems (Nuova Systems), EMC, Emulex, Finisar, Hewlett-Packard, IBM, NetApp, QLogic and Sun. Other players interested in participating in the FCoE market include Blade Network Technologies, Broadcom, Intel and Mellanox Technologies. The most obvious and most vocal proponents of this technology have been Cisco and its recently annexed incubation company, Nuova (the former Andiamo team).
Beneath the marketing gloss, FCoE looks very promising. It's certainly moving down the certification route faster than any other protocol in recent history, on track to go from inception to production in less than two years. This has been accomplished through universal backing by all participants, as well as by Nuova relinquishing its patents to FCoE with the understanding that proprietary protocols don't create a big market. Does all this unprecedented cooperation guarantee success? Probably not, but it should help to relieve any interoperability concerns.
Most new technologies aren't implemented in production environments until they're fully tested and proven. For users, the key to FCoE will be understanding its impact on data center operations. Once test and development cycles have been completed, early deployments will most likely be in the form of server fan-in environments--especially high-density blade server environments--and then move into the core from there.
This was first published in July 2008