InformationWeek

Centralizing Backup: Storage Gets Serious

Saving data is becoming a critical issue. If your company's fortunes
depend on back-up files, can you afford to take chances?

By Richard Adhikari (Issue date: July 24, 1995)

hen Northwestern Mutual Life began moving noncritical applications development off its mainframe in 1990, the insurer's information systems staff knew it was at the forefront of multiplatform, client-server computing . What the IS department didn't know, however, was that just two years later, the $48.1 billion company would be in the vanguard of yet another movement: recentralizing the storage of client-server information. At first, the Milwaukee insurer used four Emerald Systems digital audiotape units attached to the 22 Novell NetWare file servers to store business data. But the backup process started at 6 every evening and didn't end until 7 the next morning. Moreover, staff members had to change tapes manually. "The [digital tapes] were not [moving data] as fast as mainframe tapes," recalls Jeff Koeberl, a Northwestern Mutual senior systems analyst. "And there was no automation."

Problems with reliability and data accessibility have long plagued companies moving off mainframes and onto client-server environments. While distributing work and applications to more than one platform helps many business processes, it isn't always the right approach for information st orage.

That's why a number of companies have begun to recentralize storage--either in a single location on a LAN server or back on the mainframe.

"The trend is toward creating storage farms," says Dave Cappuccio, a VP at Gartner Group, an information technology advisory firm in Stamford, Conn. Storage farms help IS departments focus on maintaining the integrity of enterprise storage.

But going the LAN server route poses its own set of problems, as Fred Danback, VP of IS at NAC Reinsurance Inc. in Greenwich, Conn., learned. NAC, with revenue in excess of $450 million last year, moved applications off its IBM 4381 mainframe in 1991 to a distributed system based on IBM RS/6000, Pick Systems' Pick, and Sybase 10 platforms. The new architecture supports 10 Lotus Notes servers in seven locations. Today, the reinsurer supports more than 700 discussion databases, vacation and business travel tracking systems, corporate handbooks, organizational manuals, and telephone lists--all operating over Lotus Notes . Naturally, the amount of data that needed backing up grew by leaps and bounds.

NAC was following a classic, distributed backup model. Remote offices backed up data on 4-mm digital tape on their own servers. That information was replicated onto servers at headquarters on 4-mm and 8-mm tape systems.

In spite of this two-tiered approach, information was lost. A user accidentally erased some files in 1992. The IS staff couldn't restore the information from the previous night's backup tape because the files had been corrupted. It turned out the problem extended to tapes dating back two months. "Luckily, it was only a couple of files," says Danback. "If we'd lost the whole system, we'd have lost the business."

To fix the problem, Danback borrowed a concept from the mainframe world: automatic tape verification. This process uses backup software to confirm that data has been backed up. The technology helps businesses compare the sizes of the original file and the backup file, and flags a backup log for further action if there is a discrepancy.

Verification is offered by all the backup software providers NAC uses. These include Network Archivist from Palindrome for NAC's NetWare servers, Backup Exec from Arcada Software for NAC's Notes servers, and the backup functionality inherent to IBM's AIX that runs on RS/6000s. "It's important to keep mainframe disciplines," says Danback, who now has mainframe-style, offsite backup, archiving, and disaster-recovery systems in place.

Going The Distance
Since mainframe disciplines are so important, why not go the distance and leverage the existing corporate mainframe? About 30% of IS managers at large corporations are doing just that, says Gartner's Cappuccio.

The difference between mainframe and distributed storage comes down to the nature of the tape media. Manufacturers of 4-mm digital tapes and 8-mm tapes--the most widely used backup media in client-server environments--use helical scan technology. There, the tape is wrapped partially around a rotating drum with multiple heads. Information is written as a series of diagonal stripes across the tape, maximizing capacity in a small space.

Size is important in the client-server world, where space is at a premium. But there's a problem: Because of friction, the head tapes tend to deteriorate over time. That can mean lost data.

Linear Scan
Mainframe tapes--including increasingly popular mini-cartridges--use a different approach with data written along the length of the tape. With this technology, called linear scan, the tape and the drive heads do not touch.

Linear scan makes for lower-capacity but more reliable tapes. Here, size is not as important, as tapes tend to be a half-inch wide. And the devices holding the tapes, such as Storage Technology's PowderHorn silos, can be massive, holding up to 6,000 tapes. A company can even daisy-chain 16 PowderHorns together into a massive storage cluster.

The Online Computer Library Center (OCLC), a not-for-profit service that provides 18,000 libraries with online reference and cataloging services around the world, uses five PowderHorns at its Dublin, Ohio, location. Subscribers access the services through the Internet or dial-up data lines. OCLC runs its databases on Tandem Himalayas and IBM 3090 400Js mainframes attached to the PowderHorns. Front-end access is provided through six IBM RS/6000s linked to the Tandems and IBMs over Ethernet.

The RS/6000s come with their own 8-mm tape drives, but OCLC uses this built-in storage only to store local software and quality-assurance test results. "If it's a production system, I want them to put that data onto the mainframe for security, reliability, and recoverability," says Jerry Lynch, OCLC operations division director.

Mainframe systems offer another feature much in demand among technology managers: hierarchical storage management. HSM is a time- and usage-sensitive data management method in which companies migrate informati on to various media, depending on how soon and how often it's needed. HSM has three levels of storage: primary, which is usually a magnetic disk on a computer; and secondary and tertiary levels, which use backup tape or optical storage methods. Migrating less-frequently-used data to the secondary or tertiary levels frees local storage.

In the client-server world, HSM is particularly desirable because demand for storage is growing fast. Tom Lahive, an analyst at market research firm International Data Corp. in Framingham, Mass., says sales of hard-disk storage systems for non-Unix servers will at least double every year through 1998.

Feeding The Need
CSC Logic, a Dallas subsidiary of systems integrator Computer Sciences Corp.--which manages the National Flood Insurance Plan (NFIP), a federal program for insurance companies--saw its demand for storage soar as the company migrated to a client-server architecture. "I figured the storage needed would triple or quadruple," says Eric Trimpe, a network systems analyst at CSC Logic.

CSC Logic's new setup includes 250 486 PCs running Windows or OS/2 PCs linked over 10Base-T Ethernet to five Novell NetWare 3.1.2 servers. The system's core business process is managed through a custom-built imaging application running over an Oracle7 database.

Initially, Trimpe backed up the system to HP JetStore 5000 tape drives that were built into the LAN servers. But that took 15 hours daily, and the drives soon began running out of capacity.

Instead of buying more storage, Trimpe decided to leverage NFIP's existing IBM 3090-J mainframe. He installed ArcServe from Cheyenne Software Inc. in Roslyn, N.Y., and now uses its HSM feature to back up PC hard drives onto LAN servers. These, in turn, are backed up to the mainframe's IBM 3480 tapes using Stage3 from Boole & Babbage, a software vendor in San Jose, Calif.

The entire process is automated and has trimmed CSC Logic's backup time by 20%. It has also brought an added bonus: Because Trimpe spreads the cost of mainframe tapes over the entire system, overall storage costs dropped.

HSM has been available in some client-server backup products for two years. But according to Gartner Group's Cappuccio, "Most people at the departmental level don't know what HSM is about." He adds: "Disk costs are so low that they just stick on more storage when they need it."

Back To The Center
That may work in relatively small departments. But at the enterprise level, the haphazard addition of storage can cause data-management problems. The management conundrum is being solved, in part, through the recentralization of storage.

And because many technology professionals are comfortable with HSM in traditional data-center environments, they are demanding that it be included in new client-server systems.

Today, most major storage backup software vendors--including Cheyenne, Arcada Software, Avail Systems, Legato Systems, and IBM--already offer HSM or are thinking about doing so. Even Microsoft plans to bundle HSM within Windows NT under a licensing agreement with Avail.

As technology professionals regain command of client-server data storage, they will demand more of the features they are used to in the mainframe world.

Data storage, which went from a centralized environment to the decentralized, client-server world, finally will come full circle. Centralized data management will come back into vogue.

Comments on this story?

InformationWeek http://techweb.cmp.com/iwk