Is Fibre Channel dead?
US IT consultant Joel Snyder says iSCSI will win out
By Deni Connor, Network World | Techworld | Published: 08:00, 26 March 2007
At Network World’s IT Roadmap show in Boston last week, IT consultant Joel Snyder proclaimed that Fibre Channel is dead.
I asked him why he thought that considering everything I’ve heard and everyone I talk to says it’s not.
Snyder replied: “Fibre Channel is like Token Ring or H.323. Those who want to sell it just will refuse to accept [that it’s dead], but the sales of iSCSI by people who ‘get it’ will skyrocket, while Fibre Channel will slowly decay away. In two years, people will say ‘Yeah, of course, it was obvious.’ For now, I'm sure people will get all excited about it. The only reason people will still be buying Fibre Channel is because there are still a lot of Fibre Channel devices in the pipeline. The nail in the coffin was 10G-bit Ethernet.”
So I asked him: "You are actually saying that ‘Fibre Channel is going to be dead,’ when users adopt 10G Ethernet and use iSCSI to transport their storage. I haven't seen a lot of uptake of 10G Ethernet in the enterprise - yet. But with TCP offload and Microsoft's Scalable Networking Pack, iSCSI and 10G Ethernet should be a hit. Right now though 10G Ethernet adapters are tracking at about $1,000 each, far too expensive for someone who wants to put in iSCSI because they think it is inexpensive."
Snyder responded. “No, that's not exactly what I meant. What I mean is that anyone who, today, goes out and buys Fibre Channel is making a serious mistake, and that the path of growth and innovation is going to be down the iSCSI route.
“There was a shred of argument that Fibre Channel was faster than 1G Ethernet, although I honestly don't believe that was true for most topologies. That argument disappeared for two reasons: First, vendors discovered that it was dirt cheap to stick anywhere from 4 to 10 1G interfaces on their boxes, meaning that you didn't have to have a shared medium, but could put in a direct link to small clusters and get the same effective throughput. Secondly, 10G Ethernet is easily available now, which means that if you truly are not happy with 1G, then you have the option to jump to 10G.
“In effect, there is no reason to ever buy Fibre Channel again, starting as of about today - to do so, is to lock yourself into an expensive and declining technology. Everything about Fibre Channel is expensive: the patches, the support, the switches and the adapters. Everything about iSCSI is cheap. Given that the value in the product is the stuff that sits on top of it, why would you pay $50,000 for infrastructure to support a storage-area network [SAN] when you could pay $10,000 and be just as happy and get just as much performance as you could ever want.
“Without sounding like an old man, I've seen this evolution five or 10 times in the past 25 years and you probably have too (I did see ARCnet, FDDI and Token Ring decline), as initial technologies had some insane momentum (think FDDI, Token Ring, H.323), but then absolutely refused to step aside when a cheaper simpler way of doing things came out (think Fast Ethernet, think SIP [Session Initiation Protocol]).It is an irrefutable law of technology that simpler and cheaper win, and this is especially true when simpler-and-cheaper are going to be indistinguishable from more-complex-and-more-expensive.”
I asked him what he suggests Network World readers should do - buy Fibre Channel or go with iSCSI?
“Those Network World readers who want to be on the power curve should be buying iSCSI,” says Snyder. “Yes, probably, there are individual situations where a single 4G Fibre Channel interface for a particular application is ‘better’ than iSCSI today. But for 90 percent of the people who want to buy SANs, at ALL levels, the right answer is iSCSI and that's because the value in the SAN is the same no matter what network interconnect you use.”