High speeds, high stakes for gigabit fiber networks

Fibre Channel is touted as an all-in-one transmission standard. But can it survive in the networking arena against Gigabit Ethernet and atm?

Stephen Hardy Editor in Chief

New phrases in the English language should elicit more skepticism than the slogan "One Size Fits All." Usually this slogan translates into "One Size Fits All Who Really Don`t Care What They Look Like." Transferred to other scenarios, the "all-in-one" description has been applied to vegetable dicers and pocket knives in modern times and to medicinal elixirs of dubious value in past centuries. Purveyors of these latter potions usually were called "snake-oil salesmen," and our forefathers treated all of them more or less the same way: They ran them out of town on a rail.

With this context in mind, one would understand if a smirk came across the face of the average information technology (IT) manager when first reading about the Fibre Channel standard. A single protocol-independent scheme for data transfer among workstations, mainframes, supercomputers, desktop computers, storage devices, and displays? Point-to-point, clustered, and networked? Local area, wide area, and backbone? At gigabit speeds?

Yeah, right.

But if one would expect such a reaction, he or she would be surprised to learn that the initial applications of Fibre Channel indicate the standard works pretty much as advertised. People even seem to like it. Which raises the question of whether Fibre Channel will become as ubiquitous in networks of all shapes and sizes as its potential would indicate. Particularly in traditional networking applications, will the standard shoulder aside such competitors as Asynchronous Transfer Mode (atm) and Gigabit Ethernet? As the optical community watches the standard`s applications grow, will it see Fibre Channel acceptance increase the demand for fiber-optic technology? People inside and outside the Fibre Channel circle tell similar stories about where the standard will make its most noticeable mark -- while reciting a familiar tale about fiber`s potential to crack a new market.

Assembly required

The Fibre Channel standard represents a response to the explosion in data transfer requirements. As the power of data processing equipment has increased, networks frequently need more bandwidth for communication among computers, servers, and storage devices than their resident technologies can effectively provide.

In 1988, the American National Standards Institute X3T9.3 committee tasked the Fibre Channel working group to tackle the problem. The group was asked to address problems such as the speed and distance constraints of then-available technology, the relatively large footprints of connectors, the limits on connectivity imposed by physical incompatibilities, and the restrictions on the number of nodes that existing protocols could accommodate. The group examined contemporary equipment and standards and noticed that several existing interfaces, such as the small computer system interface (scsi) and the high-performance parallel interface, coincidentally were due for extension of their capabilities. The group decided to combine extended capabilities with simplified architecture in the same set of standards, with the hope of achieving a high-capacity transport mechanism capable of supporting multiple protocols through the same port and across the same high-speed communications channel.

The resulting specifications produced a transport scheme in which the transmission is separated from the control protocol. This wrinkle enables Fibre Channel to support point-to-point, arbitrated loop, and switch topologies with equal effectiveness. The switched interconnection scheme, called a "fabric" in Fibre Channel terminology, certainly achieves the goal of increased node capacity; while scsi supports 16 nodes, Fibre Channel supports 16 million via a single fabric.

Meanwhile, the distance and speed extensions also have been established. Vince Melendy, a sales engineer at Chicago, IL-based Methode Electronics and a member of the ncits T11.2 group that is responsible for setting Fibre Channel`s physical layer specifications, reports that completed copper specifications call for gigabit speeds to 30 m--which is a short distance compared to the established 500-m standard on 50-micron multimode fiber and the 10-km standard of singlemode fiber. Melendy reports that the standards group is revising the 10-km singlemode specifications; this work should be completed in the first half of this year. Meanwhile, the group also plans to double the currently producible transmission rate across all media from 1.0625 to 2.125 Gbaud this year as well.

Roll `em out

As the standard rounded into form, manufacturers quickly lined up to integrate Fibre Channel into their product lines. Several organizations, including the Fibre Channel Association and the Fibre Channel Loop Community, formed to promote development and use of the standard. So, who is going to use it, and how?

According to industry sources, the standard`s future applications fall into three main areas: storage, clustered networking, and general communications networking. The sources agree on the application in which the standard will make its first and probably most significant inroads.

"Storage is probably going to be the one that will have the earliest and quickest adoption of this technology because it`s just a natural as an evolution from parallel interfaces into a serial SCSI interface," says Jay Kramer, director of business development and marketing at Unisys Corp., San Jose, CA, and a board member of the Fibre Channel Loop Community.

"I think where we`ve come to is that in the storage world, the attributes that Fibre Channel brings are very, very compelling in terms of just carrying the current capability forward," agrees Julian Elliott, manager of the high-speed optical unit of Hewlett-Packard`s Optical Communications Div. "In other words, you`ve got scsi that has a certain maximum data rate and number of devices that can be connected and so forth. And Fibre Channel carries that forward, both in terms of its first implementation, which is at a gigabit, but also [in terms of its] ability to scale upward seamlessly."

Where Fibre Channel will really take off in the storage arena, however, will be in applications where IT managers combine the standard`s storage and networking capabilities to create networked storage resources (see figure).

"What seems...to be really exciting to the IT managers at the end-user level is this concept of storage-area networking or system-area networking, which would be a hub- or a switch-centric topology that allows you to distribute your storage around your organization, around your enterprise, in a manner that is more consistent with local area networking than [is] the traditional storage interconnect, which is more point-to-point," explains Elliott. "And as storage is becoming more and more critical to the enterprise and the amount of storage that needs to be connected is increasing almost exponentially, this ability to have flexibility, to have availability to your data, to be able to interconnect it in ways that are more analogous to networking, is very attractive. And I think that almost as much as the performance, the data rate, and all of that is driving Fibre Channel. I think this promise of establishing a new paradigm of connecting storage is what`s creating a lot of the momentum for Fibre Channel in the future."

This momentum will quickly erode the dominance of scsi in storage applications. "I think we`ll see a lot more of the growth of our industry being in the Fibre Channel environment, where the scsi environment will be relatively flat, but is still a huge part of the market," predicts Kramer. "So as we look to the year 2000, many are projecting that 50% of the drives and the storage that will be sold at the higher end will be Fibre Channel. That`s a significant amount of growth between now and then."

Networked storage

The Fibre Channel Association un-veiled working demonstrations of switch fabrics and loops at two trade shows last year. Linda Reed, director of communications at Mcdata Corp., Broomfield, CT, as well as a board member of the Fibre Channel Association and chair of its marketing committee, says that loop architectures are beginning to take hold within end-user environments. Meanwhile, Peter Dougherty, marketing chair of the Fibre Channel Association`s Interoperability Special Working Interest Group, envisions that storage-attached networks will evolve much as traditional data networks did -- from loops to switched networks. Fibre Channel fabric equipment suppliers are thus configuring their equipment to support both kinds of networks, Reed says.

This bodes well for networked storage -- and raises the possibility of moving Fibre Channel technology into such other network applications as local-area and wide-area networks (lans/wans) and backbones. "The first thing that people see is it`s a network," points out Kramer when describing the Fibre Channel demonstrations. "You take a host adapter, put it into a workstation, and you`re on that Fibre Channel backbone. Now the questions come up, where does this apply, and how will people utilize it? I think it will start at looking at campus environments that are already laying fiber-optic cabling, and now you`ve got Fibre Channel speed among any client workstations that are attached to that network. So, you incur the cost up front of a separate host bus adapter to support Fibre Channel. But you`re now going at gigabit speed level, which to many types of applications becomes vital--especially when dealing with file transfers of multimedia-type information."

The ability to move multimedia traffic, particularly video, has led to early acceptance of Fibre Channel as a networking enabler within the entertainment industry, according to Kramer. "If you were to look at the video broadcast field, they started with Fibre Channel more as an interconnect architecture than as a storage interface. Now they are using it for all three. But if you went to a typical studio, where they`ve got Macintosh and NTs and Silicon Graphics stations and avid editing stations... they needed a way to interconnect all of these different clients and servers. They also needed the clustering capability between work groups and then evolved putting storage right there on that same backbone," he says. "That was one that actually evolved the reverse way, as we`re seeing other market segments that are utilizing the storage side first, and then the others will come."

Fibre Channel skeptics

Yet there appears to be significant skepticism in some quarters about Fibre Channel`s future as a major player in communications networks applications. For example, Elliott points to the competition from atm and Gigabit Ethernet, which uses physical layer elements from Fibre Channel as a foundation, as a limiting factor in Fibre Channel`s lan and backbone future.

"So we`re really having kind of a concurrent step up to gigabit speeds with, on the networking side -- to the enterprise or to the work group--it`s going to be Gigabit Ethernet or it may be atm on the backbone," explains Elliott. "And then on the storage side of the server, `talking` to the storage devices, it`s going to be Fibre Channel."

Stephen Montgomery, president of market research firm ElectroniCast Corp., in San Mateo, CA, agrees with Elliott. "atm will be the number one backbone," he says, citing the fact that atm is already well-received in the marketplace and proves synergistic as the enterprise network is connected to the atm switch likely resident at the local exchange carrier`s point of presence. "Then for certain [backbone] applications, people will choose Gigabit Ethernet or Fibre Channel."

Montgomery says that this choice will largely be a matter of personal preference, depending on how closely wedded the user is to using Ethernet technology from end to end. "The people who have migrated from 10-Mbit/sec Ethernet to 100-Mbit/sec [Fast Ethernet] feel that it`s just a logical migration to go up to 1000- Mbit/sec [Gigabit Ethernet]," he explains. "However, I think that most people don`t feel that way; they don`t see it [as a logical step]. It sounds like a logical step, but actually it`s not, because once you get to a backbone, we think it`s necessary to be able to feed off to other topologies easily. And Gigabit Ethernet is great for a workstation lan, but when we`re talking about a backbone, we think it`s beneficial to go with atm."

atm`s forecasted dominance doesn`t mean all is lost for the Fibre Channel crowd, however. "Often we do need to get to the speeds of Fibre Channel, and depending upon the application, people actually have accepted Fibre Channel very well," says Montgomery. "So we do see Fibre Channel growing -- but if it weren`t for the competition from Gigabit Ethernet, it would be growing even more."

Elliott and Kramer seem resigned to a second-tier fate for Fibre Channel in communications networking, balanced by the standard`s growth as a networked -storage tool. "Fibre Channel is not in any way targeting to be the mainstream tool of lans/wans. It just so happens that in many situations, not only does it do it well, but it`s conducive to using it for these multiple functions," says Kramer. "Customers will protect their investments in Ethernet technologies. And the Fibre Channel environment will evolve as an evolution of the scsi bus for interfaces for storage and servers and become an integral part of both clustering topologies and network-attached storage. So that once these storage arrays are network accessible, it`s what would be a perfect marriage between protecting the investments in the existing lan/wan, allowing that data to now be accessed anywhere on the network."

"The whole concept of networking and a successful networking standard or data storage standard is interoperability -- having the whole industry built up around a given standard so that everything works with everything. And it doesn`t appear that for the foreseeable future Fibre Channel is going to be that standard in the lan," concedes Elliott.

According to Dougherty, Fibre Channel was mistakenly portrayed in its infancy as a competitor to atm and Gigabit Ethernet. It really should be seen as a complementary technology -- one that will prove very effective as a server-to-server interconnect carry such traffic as Internet protocol (IP). He envisions a time when Fibre Channel will even be carried native over Synchronous Optical Network backbones.

Does "Fibre" mean fiber?

Even if Fibre Channel doesn`t conquer the lan, wan, or backbone mountain, it might seem that the concept of networked storage ought to be good news for the fiber-optics community. After all, if you distribute storage devices over potentially great distances, link them via cabling, then transmit large quantities of data at very high speeds, this would seem to be a perfect job for optical cable.

Of course, nothing is this easy. Like the desktop environment, the storage world is copper-based. And like fiber-to-the-desk, fiber-to-the-storage devices face both institutional and economic hurdles.

Thus, reports Christine Foster, Fibre Channel product manager at Elkton, MD-based W.L. Gore & Associates Inc., a company that makes copper and fiber cable assemblies and media interface adapters, while Fibre Channel in general is receiving a warm response, fiber-based applications suffer from comparatively cold treatment.

"I would say 60% to 70% of what I`ve seen has been copper-based," she reports. "Because I think the initial products ...really haven`t gone across a campus or anything like that yet. [Users are] still in the early stages, where they`re just testing out whether it will work or not. So most of the distances have been under 30 m--which you can do cost-effectively with copper."

Most manufacturers have responded accordingly, Foster says. "The strategy that we`ve taken and what we`ve seen a lot of the companies doing--like the host adapter companies and people like that--is taking a default copper-type setup on a lot of their equipment," she explains. "So using a default copper port, like putting a DB-9 header on their port, and then from there they scale to optics using a media interface adapter ...when their links get beyond 30 m."

Kramer says his company sees the same initial trend. "In the short term, we still see a high percentage of copper for several reasons," he reports. "One, organizations already have copper infrastructures. Two, to get the performance of Fibre Channel, you can still do that with copper. And it isn`t until the distances dictate optical, or the cost-effectiveness or standardization that an organization is doing toward fiber optics is completed, that fiber will start to represent a larger percentage of the market."

Yet a transition to fiber is coming, industry sources say. "I think that when [Fibre Channel] goes to the 2-Gbit speed, which the group`s starting to talk about now, the copper links are going to get really limited at that point. Then probably you`ll start off with the majority of things being optical," predicts Foster.

"It`s going to be very much mixed," says Kramer. "Within a data center, which has traditionally been copper, there will be more and more transition to optical [fiber] as the costs of optical become more and more attractive. So all of a sudden it becomes just as cost-effective to put in optical as copper."

The day of cost parity between copper and fiber is coming, says Foster. The transition from CD lasers to vertical-cavity surface-emitting lasers (vcsels) for short-wavelength applications will help significantly toward this goal. "The cost of the optical transceivers is coming down," she explains. "So it`s making it, at 30 m, almost a wash, cost-wise, between optical and copper. So I think that as the data rates increase and the vcsels [prove to] be more cost-effective than the CD lasers, then even if copper could go 30 m, it might be cheaper to do optics for almost all the links." Foster also predicts that customers will eventually demand long-wavelength devices for singlemode applications.

Channeling the future

"The initial promise and vision of Fibre Channel was as a protocol-independent data transport that could carry multiple protocols -- both storage-oriented protocols such as scsi as well as networking protocols such as IP and atm. So it was kind of the Grand Unifying Theory of data transport within an enterprise for the networking environment," says Elliott.

Whether Fibre Channel can fulfill that promise -- and muscle aside Gigabit Ethernet and atm in communications applications -- remains very much in doubt. However, at least in the storage arena, the standard appears ready to bring a new level of flexible networking to IT managers. As Elliott puts it, "The pendulum is swinging from kind of a server-centric or cpu-centric enterprise network to a data-centric enterprise network. And so the ability to configure, to manage, to add to, and to scale the data is becoming very critical. So Fibre Channel is an enabling technology that`s going to allow that to happen."

As new networks grow to meet linked-storage requirements, a role for fiber optics appears in the offing--provided the community solves the same problems it faces in extending fiber to the desk. "Because our base has been copper, companies such as Gore and others are working real hard to extend the lengths of copper. And the only way that optics is going to replace it is by really lowering the cost of the transceivers," asserts Foster. "I think that`s the biggest hurdle for optics. Otherwise, people really like optics because of less electromagnetic interference and [more] noise immunity, among other reasons."

Like the typical enterprise network environment, fiber-optic backbones with copper branches for networked-storage may be feasible in the near future. How far fiber moves off that Fibre Channel backbone depends on distance--not only from backbone to storage device, but also from the cost of the technology to the amount IT managers are willing to pay. u

Gigabit Ethernet and Fibre Channel: sibling rivalry?

It seems to make sense to compare Gigabit Ethernet and Fibre Channel the way you might compare two brothers close in age. After all, big brother Fibre Channel loaned many of its physical layer attributes to Gigabit Ethernet when the latter`s standards efforts first got under way. Like a younger brother, Gigabit Ethernet has been rushing to catch up to -- if not surpass -- Fibre Channel in terms of market acceptance and standards completion.

Certainly some people see the two transport mechanisms as competitors. "As far as competition for Fibre Channel, we think that the main competition in the next few years will actually be from Gigabit Ethernet," says Stephen Montgomery, president of ElectroniCast Corp., a market research firm in San Mateo, CA.

If the two are indeed going to be competitors (although, as the accompanying article indicates, most see the standards achieving some sort of peaceful coexistence), Gigabit Ethernet will first have to get over a few growing pains. Chief among these is its operation over 62.5-micron multimode fiber. The ieee 802.3z Gigabit Ethernet Task Force established two specifications for operation over fiber: 1000Base-SX, which covers 850-nm laser transceivers operating on 50- and 62.5-micron multimode fiber, and 1000Base-LX, which covers 1300-nm laser transceivers on multimode or singlemode fiber.

The distance limitations for 62.5-micron multimode fiber were originally set at 300 and 550 m for 1000Base-SX and 1000Base-LX, respectively. However, the task force reduced the -SX distance to 260 m and the -LX distance to 440 m last year after conclusion of what was termed a "major study" (see Lightwave, September 1997, page 1).

Subsequently, another problem cropped up at 62.5 microns. It was discovered that on certain -- but not all--62.5-micron fibers, a condition called "differential mode delay" appeared when an -LX transceiver designed for singlemode fiber was used. At first it appeared the distance limits for multimode -LX would have to be reduced further. How-ever, researchers discovered that by using a hybrid jumper cable, they could transform the singlemode -LX launch into the fiber into a multimode launch, thus skirting the delay problem. The process is called a "conditioned launch" and will be offered as an option under the -LX standard.

According to Vince Melendy, a sales engineer at Chicago, IL-based Methode Electronics and a member of the Physical Layer Group of the 802.3z Task Force as well as its Fibre Channel counterpart, a subcommittee has been established to finalize tests that would ensure users that a proper light launch has occurred when using the conditioned launch technique. The results of the subcommittee`s work will be reported when the task force meets next month. Melendy says he expects that work on the standard will be completed by the task force`s March 1998 goal.

Once the standards issue is settled, where will Gigabit Ethernet make its mark? As discussed in the accompanying article, Montgomery sees Asynchronous Transfer Mode as the backbone king -- leaving the two siblings to fight for everything else at the expense of other kinds of networking technologies. Thus, he sees Gigabit Ethernet "taking market share away from two things: a little bit out of Fibre Channel and a lot more out of Fast Ethernet."

If Montgomery is correct, the fighting between Gigabit Ethernet and Fibre Channel will be more like two brothers roughhousing than a blood feud. n