Even though 4-Gbit/sec Fibre Channel technology has made only initial inroads in the field, vendors of switches, host bus adapters (HBAs), and other transmission technology for SANs have already laid plans to deliver next-generation 8-Gbit/sec products in 2008. That means design work has already started, including discussions with the optical transceiver companies that will supply pluggable interfaces. The trick for the optical component suppliers and their customers is to meet cost targets that begin at a minimal premium over 4-Gbit/sec predecessors, with a rapid path toward near price parity. Given the fact that the 8-Gbit/sec standard isn’t even finished yet, some transceiver vendors have doubts about their abilities to meet these cost expectations.
The 2008 date is part of the Fibre Channel Industry Association’s (FCIA’s) roadmap, according to Skip Jones, chairman of the FCIA and director, planning and technology, at SAN equipment vendor QLogic Corp. (www.qlogic.com). “We’re working hard on it right now,” he reports regarding QLogic’s 8-Gbit/sec Fibre Channel (8GFC) activities.
While the majority of applications won’t be ready for 8 Gbits/sec until the 2008 target, there’s a demand for the technology now in some quarters, Jones says. “In this business we always have a core base of customers who will gobble up every megabit of bandwidth you give them,” he says, listing video post-production facilities as an example.
That said, the 2008 timeframe appears about right for the market, he believes. Since QLogic OEMs some of its switches and HBAs, it has to time its development cycles to its clients’ qualification cycles. “We hope to start those qual cycles with companies in 2008 and possibly some earlier [this] year,” he says.
Page Tagizad, director of product marketing at Emulex (www.emulex.com), which also makes Fibre Channel switches and HBAs, agrees with Jones’s timeline and assessment of market demand. “We believe [the transition to 8G Fibre Channel] will start in the second half of [this] year, with samples and early adopters,” Tagizad forecasts. He adds that he expects deployment into the field to begin in the latter part of this year or early part of 2008.
Tagizad believes that communications requirements within and among servers will continue to increase rapidly, with PCI Express pipelines doubling over the next year or so. The move to virtualization-where a server is partitioned logically to emulate multiple units-also should increase the amount of input and output SANs and computer networks must support. “Just from a hardware perspective, it looks like the industry is embracing faster I/O availability and that speaks for 8-gig Fibre Channel as well as 10-gig Ethernet,” he adds.
However, it takes more than an increasing demand for big bandwidth pipes to create a market, he says. As Tagizad puts it, an “ecosystem” must develop that includes not only bandwidth demand, but applications and backup software, plus the cooperation of switch vendors, component suppliers, and the other players necessary to build 8GFC systems and make them usable. This includes backward compatibility with at least two previous generations of technology, in this case 2- and 4-Gbit/sec hardware.
Yet the impact of these factors lessens considerably without the right price points. The Fibre Channel market is notoriously cost sensitive. Jones says that customers will undoubtedly expect the transition from 4 to 8 Gbits/sec will follow the same curve as that from 1 to 2 and from 2 to 4. According to Jones, that means an initial 30% to 40% cost premium for the first 8GFC systems versus current 4-Gbit/sec equipment, with a rapid descent to near cost parity. For optical transceiver vendors, Jones says, that means initial port costs for OEMs can’t be much higher than 20% or 30% compared to 4-Gbit/sec devices.Thus, optical transceiver vendors have their marching orders. Unfortunately, since the T11.2 group that is finalizing the 8GFC standard hasn’t finished its work yet, the optical device manufacturers aren’t completely sure what they’ll have to produce.
There are a few hints. First, the devices will come in an SFP+ form factor. “I’m not even sure there’s an alternative, to be honest with you. The straight SFP is going to have EMI issues, and nobody’s really interested in a bigger form factor that I’m aware of,” says Robert Zona, marketing director of Intel’s Optical Platform Division (www.intel.com).
According to Adam Carter, strategic marketing manager, Fiber Products Division at Avago Technologies (www.avagotech.com), the 8GFC requirements have evolved the SFP+ to the point where it will not be significantly different from a standard SFP, particularly in terms of size. Jones concurs, saying that the FCIA has asked T11.2 to create a specification for an SFP+ cage that will also accept standard SFPs in the name of backward compatibility. Carter says that at the time this article was written, the only significant difference between an SFP+ cage and one for SFPs would be some fingers added to the cage to improve EMI performance. Exactly what other aspects of the transceiver would have to be engineered to ensure backward compatibility is still a matter of discussion.
The link distances that the transceivers will have to support also remain unclear. The T11.2 group appears to have focused on two variants for multimode networks and two for singlemode. Particularly in the case of the multimode transceivers, one variant (dubbed SN; see table) will focus on links of less than 100 m on OM1 and OM2 fiber. While some would debate whether these devices will cost less than the more robust SA version, Jones says it appears clear that the SN devices will likely reach the market first. Both devices will use 850-nm VCSEL technology. Jones anticipates that these specifications will reach “technical stability” in February.
The transceiver vendors contacted for this story cited the Fibre Channel community for balancing the potential cost of the transceivers with the performance they’ll need to support. “Fibre Channel likes to have the right cost point for the right performance,” explains Carter. “And so they don’t want 300 m on multimode fiber if it’s going to do some bad things as far as cost points are concerned with relation to the previous generation. So there’s been a lot of discussion on where we think the appropriate link distances are with regards to multimode fiber, as well as making sure that performance and everything else is going to be okay-not only from the optical but also from the electrical output into the transceiver.”
The result, the Fibre Channel community hopes, will be laser performance specifications more lenient than those for, say, 10-Gigabit Ethernet. This, in turn, would lead to cheaper transceivers.
However, whether those relaxed specifications will lead to devices at the expected price points remains uncertain. “In the shorter run, it’s not clear that that by itself is enough to get us all the way down to 4-gig pricing,” says Zona of the more lenient specifications. “But it at least gives us the possibility of getting there in the long run.”
Zona’s doubts are shared by others. “Part of the challenge is what the industry is using as a benchmark is the 2- to 4-gig transition, which was about at 25% to 30% premium, moving into parity as volume hit,” comments Tom Fawcett, product line director for datacom at JDSU (www.jdsu.com). “I don’t believe we’re going to be on that same trajectory here, and I think it’s still to be determined what the final specs say, what the final products look like and cost. But I would say the most aggressive we could possibly see would be the 4-gig comparison. But I think we’re likely to see a premium over that.”
One factor each transceiver vendor must consider is whether it is better to try to boost existing 4-Gbit/sec technology or throttle down 10-Gbit/sec devices. Jones says he expects the former. Carter points out a problem with that assumption.
“If you look at the 2-gig to the 4-gig, basically the 4-gig specification was actually set from the 2-gig spec, and you just upgraded it,” he explains. “For 8 gig, they’ve actually taken the 10-Gigabit Ethernet spec as the reference. And, yes, we are changing the VCSEL specs and everything like that. [Yet] I think we still have some work to do to see if [hitting the cost target is] going to be achievable or not.”
That work is well underway. Companies such as Avago demonstrated prototype devices as early as 2005, and similar prototypes for a variety of vendors are undoubtedly in customers’ hands now. All of the transceiver sources suggest that sample devices beyond prototypes from multiple vendors should appear in the first half of this year, with the expectation that production-ready transceivers should appear in the second half. The transceiver suppliers expect that the SN variant will see the most demand.
If the transceiver vendors can hit their targets, the results should prove popular next year. “Our customers, particularly on the switch side, are very aggressive about 8 gig,” says Zona. “And it seems like 8 gig and the whole equipment upgrade cycle for other features are tied together with that timing.”
“We’re seeing the demand and the interest across the board-switches, HBAs, and so on. It’s not specific to any one application,” adds Fawcett.
“We’re getting more requests for 8 gig, we’ve seen, than we had in the past couple of speed migrations,” concludes Jones. “And that’s just because there’s not a company that isn’t almost doubling its storage requirements every year or two.”
While much of the discussion surrounding 8-Gbit/sec Fibre Channel has focused on SANs, the technology holds promise for other applications. Todd Bundy, director of enterprise business development and alliances, ADVA Optical Networks (www.advaoptical.com), says the technology might be considered in grid computing environments, particularly for disk mirroring.
Bundy’s company has found success for its WDM equipment in grid computing environments, which include requirements for communications among storage resources as well as mainframe to mainframe. The storage connections would appear to hold the most promise for emerging 8-Gbit/sec Fibre Channel technology.
“We know that IBM has come out with 4-gig FICON and we know they’ll probably come out with 8-gig FICON,” Bundy explains. “The question is will storage groups support it as a Fibre Channel 8-gig between their disk arrays? They use 2 today; they’re probably going to go to 4. They’ve tested on that already.”
In fact, ADVA Optical Networking has completed interoperability testing and qualification of both 4- and 10-Gbit/sec transmission for IBM’s Geographically Dispersed Parallel Sysplex (GDPS) Server Time Protocol for zSeries and S/390 eServer environments. The qualification positions the company to offer its Fiber Service Platform WDM equipment to extend network reach over 100 km.
Bundy is keeping an eye on 8-Gbit/sec Fibre Channel in case IBM’s customers decide they’d like an 8-Gbit/sec option between the 2- and 10-Gbit/sec. However, he adds that InfiniBand also could catch the market’s attention, particularly for mainframe-to-mainframe connections. The manageability that InfiniBand could afford would be at least one reason for network managers to move in that direction, Bundy says.