Q&A: Why do we need FCoE?

Posted on July 06, 2007

RssImageAltText

—In April, a number of vendors proposed the Fibre Channel over Ethernet (FCoE) standard, which will enable users to run Fibre Channel SAN traffic over standard Ethernet LANs. Products based on the standard aren't expected until at least 2009, but the proposed standard has already generated a lot of controversy. Proponents cite the need for a unified, or converged, Ethernet pipe, while detractors argue that FCoE is little more than an attempt on the part of Fibre Channel vendors to stall adoption of iSCSI (which also allows storage traffic to run over Ethernet).

As background for an article on FCoE that will run in the August print issue of InfoStor, we caught up with Taufik Ma, vice president of marketing in Emulex's Intelligent Network Products group. Emulex was one of the original developers behind the FCoE standard and is a strong proponent. Here are a few snippets from that conversation:

InfoStor: Between Fibre Channel and iSCSI, users would seem to have enough options for SAN protocols. Why do we need another one?

Taufik Ma: Unified, or converged, Ethernet has been in the vernacular since at least 2000. The goal is a converged Ethernet that handles LAN, SAN and clustering traffic to reduce costs and management. It's been the goal for some time, but it's never happened.

InfoStor: Why not?

Ma: Mixing storage and LAN traffic on the same fabric give users the jitters. Even with iSCSI, there are usually two separate fabrics—one for SAN traffic, one for LAN traffic. Also, most large enterprises already have Fibre Channel SANs. In those environments, supplementing, or replacing, Fibre Channel with iSCSI has for the most part not happened, in part because of all the changes that would be required.

InfoStor: What does FCoE bring to the table?

Ma: The FCoE protocol defines how you take a Fibre Channel packet and encapsulate it in an Ethernet packet in the form of a payload. An FCoE adapter, for example, would send the FCoE packet into the fabric and the switch would treat it like any other Ethernet packet. So you can attach a converged Ethernet fabric into an existing Fibre Channel SAN and hold on to your existing investments.

InfoStor: The FCoE standard was proposed in April. When do you expect the specification to be completed, and when can we expect FCoE products?

Ma: The T11 working group is meeting monthly, and we expect a ratified standard by mid-2008. We expect products in 2009, or later.

InfoStor: In terms of performance, how will FCoE stack up against iSCSI?

Ma: First, the assumption is that FCoE will run over 10Gbps Ethernet [as will iSCSI]. FCoE is a very lightweight protocol—much more so than iSCSI, which has to implement the entire TCP stack. TCP is a fairly heavyweight protocol. So FCoE will have performance benefits vs. iSCSI from that perspective. However, performance isn't the major factor. End users may be more interested in the management advantages. They'll be able to use their existing Fibre Channel infrastructure.

InfoStor: But you could make a similar argument in favor of iSCSI: Users are able to use their existing infrastructure.

Ma: The typical large enterprise has separate networking, storage and server IT departments, or silos. The server group that implemented Fibre Cannel may not know Ethernet. You could have the networking group manage iSCSI, but the storage group wouldn't be happy with that. As you move into small to medium-sized businesses (SMBs), where there aren't IT silos, iSCSI has taken root. In those environments, the argument in favor of iSCSI makes more sense.

InfoStor: So FCoE is really a play for larger enterprises, and not so much for SMBs?

Ma: Yes, for the most part. FCoE has its major value proposition in larger enterprises, where they already have Fibre Channel SANs.

InfoStor: Running storage traffic over Ethernet poses problems because it's a "lossy" network that drops packets. What about the reliability issue with FCoE?

Ma: With storage protocols, you have to avoid at all costs dropping packets in the fabric. FCoE relies on Ethernet not dropping packets. Today, Ethernet has a mechanism for "lossless" behavior, called pause frames. You could conceivably run FCoE on today's Ethernet infrastructure, but the industry is enhancing Ethernet so you can have more than one traffic class; for example, SAN and LAN classes, where FCoE SAN traffic has lossless behavior.

InfoStor: What about the cost of FCoE?

Ma: The goal is to get the economies of scale that you have with Ethernet. We expect FCoE will follow a similar cost curve as 10Gbps Ethernet.

InfoStor: What changes will be required to the existing infrastructure to run FCoE?

Ma: It will require specialized host bus adapters, which we refer to as converged network adapters. These multi-protocol adapters would have Ethernet ports, but would also include FCoE functionality. No operating system changes would be required, and you wouldn't have to rip out the existing Fibre Channel infrastructure.

InfoStor: What about changes to switches?

Ma: On the Fibre Channel side, you could use existing switches. On the Ethernet side, switch vendors would have to support FCoE.

InfoStor: So you'd have to rip and replace your Ethernet switches?

Ma: The switch vendors could implement FCoE via cards. Also, this won't be as big a change as you suggest because the assumption is that users would make the transition at the same time they're making the transition from Gigabit Ethernet to 10Gbps Ethernet. FCoE will ride the wave of convergence to 10Gbps Ethernet.


Comment and Contribute
(Maximum characters: 1200). You have
characters left.