Understanding The Storage Paradigm Shift.

We are at the beginning of a second paradigm shift in computer storage. In the beginning, storage was centralized and connected to mainframes and it was captive (Fig 1). The mainframe manufacturer, a market position dominated by IBM, sold most storage. Many storage vendors, including Memorex and Storage Technology, took aim at IBM's captive storage market with plug-compatible DASD at a lower price, but mostly were unsuccessful. EMC, with a different strategy, offering technological innovation and strong customer service, was able to take 60% of IBM's mainframe captive storage market. In this paradigm, storage was reliable, accessible, and relatively secure and it was easy to manage. EMC took those four characteristics to a new level.

In the first storage paradigm shift, storage became distributed. As departmental servers, first, Unix and, now, NT too, displaced centralized mainframe computing, storage went with the servers (Fig 2).

The leading server manufacturers such as Sun, Compaq, IBM, and HP, recognized the opportunities in the growing distributed storage business. They formed separate storage companies, ostensibly to compete for open storage, but, in reality, what they succeeded in doing was to carve out their own server-attached captive storage sales. Ironically, in the open Unix environment, server manufacturers were able to maintain an 85% share of the distributed storage market.

Distributed computing proved to be a good idea, for it enabled the widespread deployment of individualized application software, but distributing storage did not turn out to be a good idea, for it compromised three vital characteristics of centralized storage: reliability, accessibility, and security. Moreover, distributed storage became expensive and hard to manage. RAID systems were developed and deployed by server manufacturers and independent storage vendors to improve reliability, but did not address security, accessibility, and management problems.

EMC and its mainframe storage competitors such as Hitachi and StorageTek offered distributed storage users a return to the advantages of mainframe storage by adapting mainframe storage systems to the distributed environment. For enterprises, EMC offered the same set of features and functions that enabled it to succeed against IBM. EMG says it does not sell storage; it sells data availability and reliability.

However, the centralized storage offered by EMC, Hitachi, and StorageTek is limited in upward scalability to 12TB. To go beyond that capacity, additional standalone storage systems are required, limiting interoperability (Fig 3). To facilitate data sharing and to improve on the efficiency of storage systems, Network Attached Storage vendors offered special purpose servers with attached storage, operating independently of application servers to perform storage and retrieval functions. Yet, at the turn of the millennium, most data storage and retrieval operations are performed at relatively slow LAN and WAN speeds.

Don't histrihute Storage

Distributed computing still works fine. Distributed storage does not.

In the second storage paradigm shift that is now beginning, storage is becoming a utility. This paradigm shift is taking place with the consolidation of dispersed storage into high-speed Storage Area Networks that are centrally managed by the network (not by the server) and that allocate storage to users on demand.

In this paradigm shift, storage will become truly independent of the application server and monolithic storage systems will give way to scalable arrays. That will open new business opportunities for high-performance, flexible JBOD storage racks, optimized for flexibility under powerful management software embedded in the storage network rather than in monolithic storage systems or running on hosts. Yet the first generation of SANs are little more than captive server-attached storage systems connecting servers to storage, using high-speed Fibre Channel switches and bridges (Fig 4). Fibre Channel equipment incompatibility and server interoperability problems are forcing the first generation SANs to be limited to operation with homogeneous servers and a restricted set of network equipment engineered to work together.

To the departmental user, this paradigm shift will increase storage availability and reliability, speed access, and free the user from administrative tasks such as backup. To the enterprise, storage will be more efficient and less costly to acquire and manage.

Dealing With The Shift

One solution is Vicom's router technology, linking high speed interfaces with any server operating system because routers can appear as virtual drives to the server. In addition, because routers can have a built-in programmable driver, one can attach to any storage or network equipment that is compatible with Fibre Channel, SCSI, or IBM serial disk protocols. This offers forward connectivity with new equipment that will become operational in the future.

Management software, operating with routers, provides capabilities such as stand-alone instant backup and restore, 7/24 data availability, dynamic storage allocation, host zoning, and security. As disk densities increase, dynamic storage allocation should be available for multiple servers at the partial disk level.

With TCP/IP capability, routers and management systems can be accessed from remote locations. Users and OEMs can add customized functions such as backup routines and security protection for remote access to a SAN because our management software is addressable with open Script Language. Management software can be integrated with storage system software such as RAID 5 and 7 offered by server manufacturers and with third party Legato and VERITAS systems to provide rubber-meets-the-road performance.

With Storage Area Networks constructed with storage routers, these capabilities need not be replicated with each disk subsystem, but can be centralized and allocated to disk resources as required. An Instant Copy feature, for example, can be assigned to any disk drives in the network and expanded as needed.

As the price of disk storage continues to drop, security functions such as remote copy, RAID One, and disk-restore will be increasingly demanded by IT managers. With well-designed routers, those functions can be expanded and extended simply by adding JBODs.

SANs using the newest routers can connect through Fibre Channel switches to homogeneous clusters of servers who need centralized data access because those routers are compatible with any Fibre Channel switch or bridge product or they can be clustered through switches to JBODs for channel-intensive computing such as video editing.

Sam Tam is the CEO at Vicom Systems