FIBRE CHANNEL IS DEAD

These were the exact words spoken by Vice President and Research Director Drue Reeves at the Burton Group Catalyst Conference in San Diego on Wednesday, with over 2000 IT professionals in attendance. I just about fell off my chair. I couldn’t believe what I heard, and from a major industry analyst. When I first said this back in 1999 when LeftHand was nothing more than a vision, people looked at me like I was crazy.

There are two major factors that are influencing analysts toward this way of thinking. Number one is the rapid adoption of 10 Gigabit Ethernet and the fact that 10 Gig-E will be on the motherboard of your servers within the next 2 years. The second thing is the Fibre Channel Industry’s endorsement of Ethernet with Fibre Channel Over Ethernet or FCoE (see my blog post “The Beginning of a New Era in Storage” November 12, 2007). IT professionals are also starting to realize the significant cost savings of having a single network to manage.

Another conference speaker made the statement, “if you are planning on buying a traditional Fibre Channel SAN today you’re wasting your money.” So what are your choices? According to the analysts it’s iSCSI or FCoE. The only problem is that FCoE won’t be shipping for a few years and they are still trying to figure out how to develop a bunch of hacks to Ethernet to provide congestion management (the IEEE 802.1Qau), enhanced transmission selection (IEEE 802.1Qaz) and priority-based flow control, which are all handled nicely by iSCSI and TCP/IP today. The good news for iSCSI shops is that iSCSI will be able to leverage these enhancements to Ethernet as well.

The question is whether or not you should wait for FCoE or deploy iSCSI now. Let’s face it, FCoE isn’t very compelling. For one thing, iSCSI is IP routable, making it a superior technology to FCoE (not routable – no IP support). Combine this with the fact that you will have to buy all new Cisco gear and HBAs for your servers just to run FCoE, and you’ll understand why I believe that iSCSI will still be the cost leader in 2 years. Intel is promising the first FCoE software initiator, but we all know how that goes. They promised the first iSCSI initiator as well, but Microsoft shipped the first stable iSCSI initiator a few years later and stole the show. I don’t see Microsoft endorsing FCoE anytime soon, and if they do it probably won’t ship for another 2-5 years.

The Fibre Channel guys are running scared and trying to save their high margin business with FCoE. In the mean time the iSCSI market is in hyper growth and moving up market very fast with 10 Gig. With LeftHand’s clustered iSCSI storage, you can build an iSCSI SAN that outperforms Fibre Channel SANs by clustering Gig-E links together, and we have over 3000 customers that chose LeftHand over Fibre Channel to prove it. Imagine when we start clustering 10 Gig links together!

The analysts are right, there’s no future in Fibre Channel. And here’s the best news of all for IT managers – it’s never too late to switch to iSCSI.

Comments

[Vijay Sivasankaran]

> For one thing, iSCSI is IP routable, making it a superior technology to FCoE (not routable – no IP support).

I've read iSCSI user accounts that say their 1/10G IP SANs are being enabled by cut-through switches and not routers. Hence the advantage of being routable means nothing inside the data center.

DR and Offsite Backup are the only specialized apps that require iSCSI then.

Would being "IP Routable" make iFCP/FCIP superior to FC?

[Vijay Sivasankaran]

Is there merit in a "Unified Ethernet Storage" device where the device could support iSCSI as well as FCoE? Apps that want the routable feature of iSCSI can use it when they want and those servers that want microsecond delays in block traffic without much load on CPU could use FCoE?

[Nik Simpson]

You would have like my presentation from last year "Fibre Channel - Dead Technology Walking". The argument is not technical, i.e. I don't have a problem with FC in particular. It's an economic argument, i.e. 10G iSCSI and FCoE will eat into the conventional FC market substantially, further shrinking its already anemic volume. At some point FC vendors have to start looking at the cost of R&D for subsequent generations and make some tough decisions.

[John Spiers]

Hi Nik,

Thanks for your comment. I totally agree. Drue and you both made an excellent business case for the end of traditional Fibre Channel as we know it today.

BTW, great conference! The feedback I received from the attendees at the Server and Storage Virtualization track were very positive.

John

[John Spiers]

Vijay,

Yes, IP becomes important for remote replication and other SAN services outside the data center. Inside the data center things like connectivity management, server failover and security become easier to manage when on IP. Consolidating as many services as possible into an IP-based network makes good business sense. It simplifies your architecture and eliminates the headaches that come from managing disparate equipment and protocols.

Years ago they were pitching iFCP and FCIP as alternative technologies to iSCSI, but that obviously didn't happen. These protocols have many challenges associated with frame management and buffering when not running end-to-end, and are only used in gateway routers to move Fibre Channel frames over longer distances.

LeftHand outperforms many Fibre Channel solutions, not because of iSCSI being a lower latency interconnect, but because of our grid architecture that eliminates the controller head I/O bottleneck. If very low latency is a customer requirement then iSCSI with RDMA & TOE technology is a viable option as well.

John

[Drue Reeves]

Hi John

One caveat, I also said that the installed based of FC isn't going away soon, and thus, FCoE has legs as a bridge technology. But in general, you're correct about our sentiment. Ethernet-based storage -- and iSCSI in particular -- have many advantages over FC and thus, further investment in FC doesn't make much sense.

I agree that the traditional FC vendors know that Ethernet storage is inevitable, but one of the remaining questions is whether the traditional FC vendors will migrate their enterprise storage arrays to iSCSI or build an FCoE target.

I also made a strong implication that HP was looking at LHN as a potential acquisition. :-)

Drue

[John Spiers]

Thanks for your comments Drue. Yes, I agree that FCoE has potential as a bridge technology depending on cost and complexity. It may come down to whether or not a software initiator or low cost HBA is available.

John