A divide within the storage industry over how to connect storage systems to the rest of the network is growing--and leaving customers with a growing sense of unease.
On one side of the schism is Fibre Channel, an expensive and complicated but functioning and improving networking standard promoted by market leader Brocade Communications Systems. On the other side of the divide is iSCSI, a young technology that promises simpler networks and lower costs because it's based on the much more common Internet Protocol (IP).
To a certain extent, the two technologies are appropriate for different uses. But because fundamentally they do the same job, hardware and software makers as well as customers must bet one way or the other or accommodate both. Both will also be making a big push for their respective causes in 2002 with a host of products.
The choice between the two technologies comes at an awkward time, just as the idea of attaching storage systems using networks is gaining traction and setting the stage for an industry expected to grow to $16 billion a year in revenue by 2004. Those buying or making the products must decide now where to devote scant resources.
The industry has faced these knotty decisions before--for example, whether to back mainframe, Unix or Windows servers, or whether to wire business with copper cables or fiber-optic lines.
"It's another branch in the decision tree," said Bill Mottran, Compaq Computer's director of marketing for storage products. For Compaq, "it does mean having two sets of expertise."
Now the stakes are getting higher, as companies with interests in the two technologies advance their plans.
Fibre Channel transfers data at the rate of 1 gigabit per second (gbps), but the change to 2gbps is under way. Brocade has begun shipping its 2gbps Silkworm 3800 switch. And rival QLogic, which is pushing lower-cost Fibre Channel networking equipment to try to broaden the market, has begun developing 10gbps Fibre Channel chips it says will be done by the end of 2002.
The iSCSI standard, meanwhile, is passing a critical step as companies such as Adaptec and Alacritech have begun putting finishing touches on dedicated chips that will make iSCSI fast enough to compete with Fibre Channel. Adaptec showed cards that connect servers to iSCSI networks at the Comdex trade show and says it will ship products by mid-2002.
But the prospect of two independent networks leaves some cold.
Conventional wisdom is that customers will lean toward iSCSI for lower-end tasks or when spanning long distances that Fibre Channel can't manage. Fibre Channel will remain useful when faster transfer speeds are required, especially in "data centers" that house large companies' collections of computing equipment.
"If you're going to do Fibre Channel, how do you make it cost-effective?" asked Illuminata analyst Jonathan Webster. "If you're going to do iSCSI, which has a cost advantage, how are you going to get the performance?"
The questions are important, as the computing industry has elevated storage devices from a lowly subsystem of server networks into independent products with their own features and software. The change helped large companies embrace a design called storage area networks (SANs), in which servers use a special-purpose storage network to read and write data on storage devices.
SANs are a solid business, with Fibre Channel networking components comprising a $2 billion market this year and expected to grow to $16 billion in 2004, according to market researcher IDC.
But SANs based on Fibre Channel have been tormented by interoperability problems; different companies' switches often don't cooperate. And having to choose between iSCSI and Fibre Channel will mean even more interoperability challenges.
While immature compared with Fibre Channel, iSCSI has the backing of three huge computing companies: Cisco Systems, Intel and IBM.
"There are three gorillas moving around in the iSCSI market," Webster said. "You don't have those gorillas moving around in Fibre Channel."
Those and others are working to complete the first version of the standard by the end of the year, with the Internet Engineering Task Force expected to ratify it in the first half of 2002.
Key to the technology is the ability to speed up IP--specifically, the TCP/IP processing that packages data to be sent on the network. EMC, the top maker of high-end storage systems, has said the arrival of special-purpose chips to handle TCP/IP will be the key moment when IP storage becomes workable.
"One of the key technology challenges is to offload a lot of the IP processing from the server CPU," said A.G. Edwards analyst Shebly Seyrafi. "When that happens, it's a possibility over time that IP storage gradually scales up to the data center."
Accelerator chips are key to IBM's avid boosting of iSCSI. "We expect by mid-next year there will be iSCSI implementations that are comparable in performance to 1gbps Fibre Channel," predicted Claude Barrera, director of technology strategy for IBM's Storage Systems Group.
But IP acceleration chips afflict IP storage with some of the price tag problems Fibre Channel suffers. "The accelerators are very compelling, but what's the cost factor?" said Giga Information Group analyst Colin Rankine. "If it adds $200 or $300 to the cost of a card, it's going to slow the adoption rate."
Though iSCSI won't always directly compete with Fibre Channel, the technologies will overlap in the market, said Compaq's Mottran. "As the pricing pressure is applied by iSCSI, there will be a movement" to lower Fibre Channel prices.
Fibre Channel: Here and now
One word sums up what Fibre Channel has going for it: incumbency. Existing products, software and customers give the technology a built-in advantage.
IBM, though pushing iSCSI hard, isn't deluded about what technology prevails today. "For the present and next several years, the Fibre Channel solution is going to be the more complete one, in particular because of the support software," said Barrera, who also is secretary for the Storage Networking Industry Association trade group.
The move to 2gbps Fibre Channel should be complete by mid-2002, he said. "The step to 2-gigabit is well under way, and it looks fairly smooth," he said.
Brocade, which Gartner Dataquest analyst James Opfer estimates has 62 percent of revenue in the market for Fibre Channel switches, has begun its changeover to 2gbps Fibre Channel. It plans to ship the high-end 128-port Silkworm 12000 switch to support the higher speed by the end of the year. And about the time iSCSI will be maturing, Brocade expects to have 10gbps products.
"The time frame for 10gbps is 2003," said Steve Beer, director of product marketing for high-volume platforms at Brocade. High-speed IP storage, meanwhile, will be burdened with high costs as well, Brocade believes.
Nevertheless, Brocade is hedging its bets, saying its high-end switch will support iSCSI connections as well. "The timeline we've mapped out is early deployment around 2003 and ramp-up in 2004 and 2005," said Camden Ford, Brocade's senior product marketing manager.
One key to the future of Fibre Channel is whether its advocates cede the lower end of the market to iSCSI or put up a fight. QLogic, which sells Fibre Channel chips but has struggled to dent Brocade's dominance, is willing to fight where profit margins aren't so plump.
The company has just begun selling a $9,999 collection of server cards and a switch for a small network in response to one of the primary reasons customers avoid Fibre Channel: cost. QLogic's sales partners have learned to not even approach a SAN customer unless they have $250,000 to spend, said Frank Berry, QLogic's vice president of marketing.
QLogic believes its products will help keep Fibre Channel from being relegated to a niche product and will prove out IDC's $16 billion forecast. "You can't make this a $16 billion market if your entry point is $250,000," Berry said.