Goodbye Fibre Channel, hello InfiniBand?

Posted on February 02, 2000

By Richard R. Lee

Over the past couple years, the nascent Fibre Channel storage area network (SAN) industry has grown from an idea (not necessarily a new one) to a multi-billion dollar market opportunity. Some analysts predict the market will grow to more than $20 billion by 2005.

Click here to enlarge image

A number of innovative venture-capital-funded startups have emerged as leaders in this new technology. Their success is being fueled by the use of a core technology (Fibre Channel) that is both widely supported and based on standards, and by the tremendous pent-up demand within the IT community for a radical new approach to storage and its day-to-day management. To my knowledge, this type of dynamic has never occurred in the storage industry's 40+ years.

In light of these developments, why do Intel, Microsoft, and several of the larger platform vendors remain in the shadows? Each of these companies has a significant vested interest in seeing the challenges associated with enterprise storage tamed by an affordable and easily deployable solution. With the recent unveiling of the InfiniBand trade association and its vision for the future, it may now be clear why these virtual monopolists have remained so invisible until now.

Fibre Channel-based fabrics have for some time been touted as the cure-all for bandwidth, scalability, and reliability issues that have plagued enterprise storage systems for decades. The vendor community has even been positioning these fabrics as the future backbones for all types of advanced clustering and IP traffic. As advertised, these fabrics would allow for true multi-node, scalable clustering, as well as high-bandwidth inter-node communications requirements using the virtual interface architecture (VIA).

Click here to enlarge image

Additionally, the Fibre Channel community envisions that these fabrics will interconnect with IP-based public networks to form wide-area SAN environments. This posturing has been in concert with the Fibre Channel community's insistence all along that Fibre Channel is the only network that can provide high bandwidth and low latency for all types of application environments.

The reality, however, is somewhat different. For the past two years or so, the founders of the InfiniBand trade association engaged in a public battle to determine who was going to own the intellectual property associated with the server industry's next-generation I/O standard. This battle pitted the NGIO camp against the FIO camp. After wasting a couple of years and millions of dollars fighting this battle, the two organizations finally merged late last year under the name System I/O-and, more recently, InfiniBand-at which time they unveiled their plans for market domination.

If successful, the InfiniBand community could derail the Fibre Channel community by 2003 or so. The InfiniBand group's plans call for the creation of a ubiquitous switched fabric that is directly connected to each node via a direct memory access scheme that is CPU- and PCI-independent. This switched fabric will be able to support multiple classes and types of I/O and IP services simultaneously.

These capabilities, if deliverable, could be a show-stopper for the Fibre Channel community. Not only will InfiniBand solve systems-level problems that continue to exist in Fibre Channel-based clusters and IP environments, but it will perform at higher levels, scale infinitely, and cost a fraction of what the equivalent Fibre Channel-based solution would cost. These are compelling reasons to question the future of Fibre Channel and the companies who have dominant positions in this market today.

If the potential benefits are so compelling, then why is all of this so bad? The reason lies in the fact that most organizations have already spent several years in a "wait-and-see" mode to determine if Fibre Channel is really mature and viable enough before taking a leap of faith and migrating their entire storage infrastructures. Many Fibre Channel deployments are now in the early stages, and most organizations are highly reticent to abandon the devil they know for the devil they don't know.

Neither Fibre Channel nor InfiniBand will ever reach critical mass in the market if IT decision makers perceive that the two camps are not only at war, but are mutually exclusive in terms of technology and direction. Perceptions along these lines can only serve to harm the entire storage industry and to stifle growth and innovation at a time when it is sorely needed.

We are all aware of the tactics used by the monopolists in the computer marketplace. In spite of advertising themselves as friends to the end-user community, the core motivation of these companies is to lock users into their technology and intellectual property to gain continuous revenue streams from new licenses and upgrades. The thought that the monopolist companies would play nice in the SAN sandbox with the smaller innovators who are currently steering the direction of Fibre Channel is nonsensical when viewed in historical perspective.

To eliminate the growing potential for chaos and confusion, the IT community needs to send a loud wake-up call to all of the players in the Fibre Channel and InfiniBand camps. The message: no tolerance for any type of "winner-takes-all" battles for market dominance. Every responsible IT manager should make the position clear that they are not going to support either community to any significant level unless the two groups become unified and move down a common path.

Richard R. Lee is president of Data Storage Technologies Inc., a management and technology consulting firm in Ridgewood, NJ. He is a frequent contributor to InfoStor, and the author of "Windows NT - Microsoft Cluster Server" (Osborne/McGraw-Hill) and the forthcoming "Network Storage Solutions - SAN and NAS Design and Deployment" (Macmillan Technical Publications). He can be contacted at dstnj@bellatlantic.net or (201) 251-6620.