A debate hovers over the appropriate use of SCSI, Fibre Channel, and Ethernet technologies in the server and workstation storage subsystem. Unfortunately, the respective proponents of those technologies are applying them to an overlapping set of storage problems with equal vigor. Each problem, however, has varied demands and deserves its own solutions.
The advantages of storage-area networks (SANs) include improving the backup capability, manageability, reliability, and availability of storage. In an online environment, where storage capacity has doubled every year even though hard-disk-drive prices keep falling, the total cost of storage ownership has skyrocketed. Storage networking can resolve this problem, but only if the solutions are applied intelligently.
Three distinct situations must be addressed. First, hard drives must be connected to systems. Then, local compute systems must be joined to local storage systems. And those local storage systems need to be tied into remote storage systems. Very specific issues drive each of these forces.
Typical servers and workstations use between two and ten drives, directly attached to a motherboard ASIC or a PCI add-on card. Customers are seeking high reliability, end-user upgradability, and compatibility with a wide variety of devices. Joining hard drives and other devices to servers demands an interoperable solution with the ability to aggregate multiple devices onto one bus. The recently approved Ultra160 SCSI standard provides one solution. It deals with the long history of legacy compatibility and provides a clear roadmap for future performance gains. The bus architecture is the simplest, lowest-cost solution for mainstream disk attach.
At the subsystem interconnect level, SANs demand technology which can provide the low latency and overhead of SCSI, but in more complex topologies. Star and loop configurations are the most common in SAN environments. Fibre Channel's native serial cabling and hub support make it ideal for connecting one or more storage subsystems to multiple computer servers.
Fibre Channel also has been promoted for the drive-attach solution within the storage subsystem. Some high-end configurations, like hundreds of drives, begin to look like systems unto themselves and can take advantage of Fibre Channel's interesting topologies. But there's no reason to burden mainstream servers and workstations with the standard's added complexity. And, while SCSI is the major vehicle to implement two-node clustering using multiple initiators, that topology really shouldn't be pushed to the complex, high-node-count solutions addressed by Fibre Channel.
Longer-distance storage connections are required for campus-level SANs, remote disaster-tolerance solutions, and database-replication problems. Again, Fibre Channel and bridged Fibre Channel have been proposed for these solutions. But the idea of using either to re-cable a physical plant counteracts the effort to lower the total cost of storage ownership.
The industry's challenge is to set aside any dogmatic views about the use of these technologies and search out the appropriate match between the problem and the solution. By allocating resources to resolve these issues in an integrated fashion, the industry will be able to make storage networking into a mainstream technology. Unless vendors and standards bodies take a more balanced approach, the storage I/O market will remain confused about when to use what.