The new technology isn't ready for the big leagues just yet
Carlson Shared Services can likely count itself among a rare group. It has taken a Hitachi storage subsystem, an IBM OS390 mainframe, 20 servers and 12 Nishan Systems switches to build the company's first IP-based SAN with a 10-TB capacity.
Any questions Carlson executives had about performance were put to rest: It took 56 percent less time to run core Oracle database financials in June. "Twenty-five people worked on this," says Gary Johnson, architecture consultant at Carlson Shared Services, part of Minneapolis-based Carlson Companies, which provides services to the hospitality, travel and marketing industries. "We did a lot of testing to be sure the performance worked fine."
In the past two years, IP storage has been blessed with a lot of fawning publicity. But, so far, Carlson's IP SAN is one of only a handful of examples of a company actually using IP storage technology. That's because the tried-and-true method of IP, the most widely used network protocol in the world,is encountering some of its own bumps and bruises in the storage world as the hype dies down and sobriety sets in.
The mood among industry experts now is that IP storage may not be ready for the big leagues yet. And some are suggesting that, for now, its place is in small-scale deployments, such as the SMB market or workgroup and departmental implementations.
Questions swirling around IP storage range from when the iSCSI standard will be ratified, to how the crucial TCP offload engines (TOEs) should be designed, to whether IP storage is more likely to be used for small-scale deployments. In June, reports surfaced that IBM had quietly stopped selling its TotalStorage IP Storage 200i subsystem. Ironically, Julian Satran, one of the company's Israel-based engineers, is the lead writer of the iSCSI standard. Moreover, Mountain View, Calif.-based 3Ware officially discontinued its PalaceAid iSCSI subsystem because, according to 3Ware spokesman Mike Wentz, it required a lot of hand-holding and had performance issues.
"Am I disappointed in the way iSCSI is turning out? Only if I were an idealist, because it was capable of so much more," says Jon William Tiogo, an independent consultant and author of The Holy Grail of Data Storage and Management.
"I had hoped iSCSI was going to be implemented cleanly%85 [but it remains to be seen whether iSCSI is going to play out any differently than Fibre Channel."
What Tiogo is referring to is the interoperability hurdles that stymied Fibre Channel. In fact, according to Tom Clark, director of technical marketing at Nishan and a noted Fibre Channel expert, SANs have only penetrated 15 to 18 percent of the marketplace. A glaring example of this interoperability problem happening to IP storage is the debate between Alacritech and Adaptec about TOE designs. The most crucial part needed to make IP storage as fast as Fibre Channel are the TOEs, or accelerator cards, which address a basic problem when using IP to transport storage data.
IP depends on the TCP transport layer to move data packets in order, regardless of how jumbled they have become en route. That burdens the server with more I/O overhead than typically found in Fibre Channel, which was designed specifically to move block-level storage data. By putting the TCP/IP processing into hardware, you have more reliable data transfers without putting an additional burden on the server's CPUs. "I have said this from the beginning: Without TOEs, there is no iSCSI," says Arun Taneja, an analyst at Enterprise Storage Group, Milford, Mass.
But Alacritech, San Jose, Calif., and Adaptec, Milpitas, Calif., disagree on how the accelerator cards should be designed. Alacritech, which has its roots in the networking world, favors a partial TCP offload,the TCP data transfers are done in hardware while the connection process is via software. However, Adaptec, with its heritage in storage, believes in a full TCP off-load,so the entire TCP/IP process takes place in hardware.
"This could mean [IP storage is headed for the same interoperability issues [that exist in Fibre Channel," says Tom Hammond-Doel, a Fibre Channel Industry Association (FCIA) board member. "Plus, I'm starting to hear that the performance they are getting out of the TOEs is not quite what they expected."
The evangelists' vision of using IP from end-to-end in SANs has been cut back. Most agree that Fibre Channel will continue to be the dominant technology at the back end of storage subsystems (meaning from the controller to the disk drives). That portion is destined to remain Fibre Channel- or SCSI-based, largely because more vendors working on IP storage are concentrating on devices for the network and server side of the equation.
"I don't see iSCSI residing in the back end of the controller," Hammond-Doel says. "Nothing is going to stop iSCSI. The questions are: Where is it going to reside, and is it going to be complementary to Fibre Channel or competitive with it?"
When Carlson Shared Services wanted to implement a SAN, it looked to start-up Nishan Systems,which is intent on being the Swiss Army knife of IP storage switches,because its products support iSCSI, IFCP and FCIP. Ultimately, Carlson and Nishan set up a system that used IP and Fibre Channel as complementary technologies for a SAN designed to handle between 30 percent to 35 percent annual capacity growth, a condition Carlson has been experiencing.
The team replaced five EMC Symmetrix subsystems with one HP XP512 disk array, a unit HP resells from Hitachi. At the core of the infrastructure are Cisco's IP/Ethernet Catalyst switches. But Nishan's 4000 series switches sit on the front end of the Sun Solaris servers and the disk array, basically doing the Fibre Channel-to-IP conversion. "Fibre Channel does a fine job, but it doesn't have the scalability. You won't find many Fibre Channel SANs that have more than eight switches," Carlson's Johnson says. "The advantage of IP is you can segment the network into logical separate networks. You can't do that with Fibre Channel."
Throughout this entire iSCSI debate, it seems like Nishan is trying to maintain its focus on making IP storage an enterprise-class technology. Unlike IBM's 200i device, Nishan executives say they have tried to set the bar higher with their first product, a multiprotocol switch that works at wire speed and debuted in February 2001. They are trying not to get distracted by the political wrangling within the Internet Engineering Task Force that has delayed the release of the iSCSI standard at least three times, with the next date set for sometime this month.
"We are trying hard to get people focused on solutions," says Gary Orenstein, director of marketing at Nishan.
The company has announced several customer deployments besides Carlson, including Wells Fargo in conjunction with Hitachi Data Systems, J.P. Morgan in conjunction with XioTech, and Mesaba/Northwest, a regional company of Northwest Airlines.
Nishan's Clark succinctly measured interest in IP storage: "They are not quite beating down the door, but they are knocking aggressively," he says. "We also have had a lot of interest with partner companies, as well as with the channel. People are beginning to understand."
Such steely focus can't be said of the alliances created among veteran vendors during the early days of iSCSI. Most have broken down. For instance, the pack between Cisco Systems and Brocade Communications, formed more than two years ago, reportedly has fallen through. In a recent earnings call, Brocade executives conceded the two companies decided to pursue different strategies. Many say no love is lost between them: The pact between the IP switch king and the Fibre Channel switch top dog was an unholy alliance from its inception.
"Cisco wanted a universal FCIP protocol to connect to anybody's switches," Tiogo says. "But Brocade created a protocol that would only work with Brocade switches."
But that only highlights a more troubling issue: No longer is there a single company with a stick big enough to wave over everybody's head to force compliance with a single protocol. And, Tiogo says, without a dominant player, iSCSI becomes just as balkanized and open to as many interpretations and different implementations as is Fibre Channel.
The Sept. 11 tragedy may also have affected the IP storage adoption rate. FCIA members say the adoption rate of 2-Gbps Fibre Channel "has totally taken off." In fact, switch manufacturer McData reported during its second quarter earnings that demand for 2-Gbps products was significantly stronger than expected. The company's switch revenue was up 80 percent, largely related to its 2-Gbps launch.
"When 9/11 happened, people said, 'Oh, my God, my data is valuable!' People started to fall back on their conservative instincts," Hammond-Doel says.
The death of Fibre Channel was greatly exaggerated. But now, some are doubting the early estimates of actual IP storage adoption in 2003. Nishan is the sole voice talking about actual IP storage deployments. "I think there is a lot of fast talk around iSCSI right now," Tiogo says. "And they learned it from the fastest talkers of them all, the FCIA%85I think you are going to be well-served to experiment with [iSCSI devices as they become available. You can deploy [the technology in a small context, but don't bet the ranch on it."