By Jon William Toigo
For the past two years, the storage area network (SAN) has been the darling architecture of the storage industry and the trade press. You would have had to have been marooned on a desert island not to have heard about the promise of SANs-how they will simplify data storage allocation, provide unlimited storage scalability without downtime, and ease the storage management burden while reducing management costs. In short, SANs are described as a panacea for just about everything that is ailing corporate IT.
Early on, SANs became associated with the Fibre Channel protocol. Fibre Channel, a decade-old serial interconnect originally designed for server clustering, provides a low-latency, gigabit-speed connection for storage devices, and is said to be superior to parallel SCSI in many ways. With Fibre Channel, storage devices can be placed farther away from servers than is possible with SCSI. Plus, the interconnect offers an order-of-magnitude improvement over SCSI in terms of the number of devices that can be connected to a loop or switched fabric configuration.
Effective marketing has positioned Fibre Channel as a synonym for SAN. Only with Fibre Channel interconnects and fabric switches, advocates argue, can an efficient back-end network for storage be created that will enable universal access by servers and end users.
The widespread adoption of this view has created overnight fortunes for manufacturers of Fibre Channel devices, such as swit ches, host bus adapters (HBAs), and SCSI-to-Fibre Channel bridges/routers, and sent more than a few SCSI disk drive and HBA manufacturers scrambling to equip their products with the latest interface.
While recent discussions of InfiniBand have suggested a potential competitor to Fibre Channel as the dominant SAN interconnect (see February's Opinion), the industry on the whole has fallen into lockstep with the "Fibre Channel equals SAN" hypothesis.
Against this backdrop, however, another competitor to Fibre Channel SANs has been quietly emerging. It first came to my attention during an IBM-sponsored SAN webcast last fall. During the panel discussion, analysts from International Data Corp. and Gartner Group seemed deliberately vague in their definition of what constituted a SAN. During the follow-up Q&A session, several participants suggested that SANs did not even require Fibre Channel.
One participant pointed out that his company had established a SAN "as a sub- network" of its enterprise Gigabit Ethernet LAN. He wanted to know why IP and Gigabit Ethernet, familiar and ubiquitous technologies in many IT settings, were being overlooked in SAN discussions. A satisfactory answer could not be provided on-line, and an email response was promised.
The question was underscored several weeks later during an interview with a chip manufacturer. The representative of the company, which builds ASICs and other devices for well-known vendors of LAN switches and network interface cards, casually observed that Fibre Channel would not be the dominant SAN interconnect in the future-at least if chip orders and board designs were any indication.
He observed that Cisco and other prominent LAN switch vendors were not about to allow upstarts in the Fibre Channel switching world to encroach on their turf. SANs are, first and foremost, networks. If SANs catch hold in the market, he argued, the name-brand network equipment vendors would step in and swallow up the industry.
Though long on bravado and short on quotable facts, his statements made sense. This phenomenon occurred with the advent of "IP switching." The technology was introduced, made popular, and franchised by a little-known firm in the early 1990s. However, the innovative vendor's fortunes were cut short when IP switching suddenly appeared as an integral function of the operating software of mainstream LAN switches. In the end, the originator of the idea simply disappeared from the marketplace. The same thing could happen with Fibre Channel SANs.
Over time, this view was confirmed by sources within the development shops of several leading storage product vendors. Managers confided, off the record, that their companies' formal, public endorsement of Fibre Channel did not necessarily reflect the views of their product engineering departments. Fibre Channel SANs were "marketecture" rather than architecture. However, open criticism of the Fibre Channel-equals-SAN equation was politically dangerous. Instead of going on record with a SAN alternative based on Gigabit Ethernet and IP, managers preferred to tow the company line.
Developments last month suggest that the situation is finally changing. On March 7, Merrill Lynch reported that storage array leader EMC was preparing to introduce support for the Internet Protocol (IP) as a basis for storage networking. The result of the move, which would likely be followed by other industry leaders according to the securities firm, could relegate Fibre Channel "to a niche."
The following week, information surfaced that IBM and Cisco were on the verge of delivering to the Internet Engineering Task Force (IETF) a proposal for a SCSI over IP standard. This Request for Comment (RFC) signals the advent of IP SANs that will use familiar Gigabit Ethernet (and its successors) as a SAN interconnect.
The fact is that today's Fibre Channel SANs fail to deliver their business value proposition. Interoperability issues persist among equipment vendors. Cables are expensive, and must be custom-built if connecting products from different manufacturers. IT staff training in the new network protocol is costly and time-consuming. And alternative storage architectures, such as IP-based network-attached storage (NAS), provide tactical solutions that are easier to deploy and less costly to maintain than Fibre Channel SANs.
Fibre Channel advocates point out that these deficiencies are common to all interconnect technologies in their first stages of evolution. Even Ethernet suffered from incompatibilities early on. Ultimately, they insist, Gigabit Ethernet, with its IP stack, is not a viable transport for moving data in a storage infrastructure.
This argument flies in the face of one important fact: Fibre Channel does not provide a means for in-band storage management. A separate network, typ i cally Ethernet-based, must be established among the storage network components to provide management. Current attempts to bring Fibre Channel SAN management in-band involve the use of an IP stack operating over the Fibre Channel interconnect. Doing so obviates the bandwidth and latency advantages Fibre Channel is said to hold over Gigabit Ethernet.
Bottom line: If Gigabit Ethernet and Fibre Channel offer comparable "speeds and feeds" as SAN interconnects, Gigabit Ethernet or IP SANs will win the day given their cost advantages over Fibre Channel SANs and their greater deployment efficiencies.
This message has not been missed by storage vendors such as Network Appliance and CrosStor Software, which are already looking to leverage their successful NAS products as "portals" for IP-based SANs. The message also delights companies like NetConvergence, which offers a "universal storage gateway" that enables heterogeneous storage interconnects to interoperate seamlessly.
And, of course, the message should come as no surprise to Gigabit Ethernet switch vendors like Cisco who, while showing little interest in storage per se, have demonstrated unswerving dedication to ensuring that their products dominate all technologies that involve networking.
Fibre Channel may persist for some time as a channel technology endorsed by storage vendors. However, once standards emerge for SANs based on packetized SCSI over IP and Gigabit Ethernet, the promises of SANs may finally be realized.
Jon William Toigo is an independent consultant and author based in Dunedin, FL. He is the author of (Prentice Hall PTR: 2000), one of eight works covering business automation issues. He writes regularly for the trade press and assists technology companies in identifying and articulating the business value proposition of their products and services. He can be contacted at (727) 736-5367 or via email at firstname.lastname@example.org.
Jon William Toigo