<B>Special Feature:</B> Calling for NT Backup

As NT's presence in the enterprise data center increases, "mainframe-class" backup is becoming a goal for system backup vendors. But can NT backup ever become "mainframe-class"? What will it take?

In a fortunate error of judgment, I referred to Windows NT backup as entering the "mainframe-class" of applications in a networking column I wrote last year. Shortly thereafter, a DASD administrator at a large IBM Corp. mainframe shop e-mailed to tell me that I obviously knew nothing about mainframe backup, rattling off a long string of data management utilities that would never make their way into the NT world, as far as he was concerned.

However, despite its seemingly immature nature, Windows NT backup is currently generating a lot of excitement, with new vendors entering the market, bringing many of these capabilities that once were unique to the mainframe world -- and more. The industry is quickly developing tools that provide better support for distributed systems backup, improved manageability, and support for intelligent storage techniques such as hierarchical storage management. The NT market may even have support for emerging technologies, such as storage-area networks (SAN), before support for these are even available for the mainframe space.

Two Roads Converge

Windows NT backup users are best characterized as two converging forks in the road. On one fork, enterprise customers are downscaling to achieve the promise of ease of use and low cost. On the other fork are the so-called "upscalers," named that because they were raised on Windows 3.x. "The ultimate challenge for any software or hardware vendor who wants to play in the NT market will be to achieve an optimum balance [between the two markets]," says Lisa Haut-Mikkelsen, ADSM product manager for IBM’s storage systems division.

The upscaler term in this case does not refer to single Intel desktops. It means the Intel world in general. So, NetWare, NT and all the "early" PC network platforms would qualify. Vendors such as Seagate Software (Scotts Valley, Calif., www.seagatesoftware.com/homepage), Cheyenne (now part of Computer Associates Int’l Inc.) and Stac Inc. (San Diego, www.stac.com) would all be considered coming from the "PC market" as opposed to the "legacy" backup vendors, which made the mainframe, Unix, VMS, AS/400 and "non PC" platforms their focus.

Currently, the "PC" vendors have the lion’s share of the NT backup market. The legacy vendors are just getting into the NT market.

However, mainframe computing has influenced Windows NT in a variety of ways, and this influence is becoming increasingly evident in the area of storage management and backup. While NT backups are notorious for their unreliability, the mainframe world has been doing reliable backups for years.

"Mainframe guys invented storage management: backup/restore, HSM, etc. Many of the basic concepts of backup planning and tape management come from that side of the house," observes Dave Ballard, group marketing manager for NT products for Veritas Software (Mountain View, Calif., www.veritas.com).

Companies that earned their stripes in other high-end markets are, in turn, making a mark on Windows NT backup by reusing the lessons they have learned. For example, Veritas’ background as a major supplier of volume and storage management for Unix systems gives the company a running start to become a contender for NT backup products.

However, other vendors maintain that there are fundamental differences between NT backup and mainframe systems. According to Jonathan Greene, director of storage product management with Computer Associates, a vendor that provides products for both markets, the key difference between mainframe or Unix backup and NT backup needs still lies in the quantity of data that mainframes must handle as opposed to Windows NT systems.

Greene points out that backup and management of the huge volume of data in mainframe environments requires extremely efficient and reliable data movement and media management. Windows NT, by contrast, grew up in a world of virtual anarchy. Data is often an order of magnitude smaller than the mainframe but is distributed across many systems, not centralized.

The bottom line is that backup in a Windows NT environment still relies on a rather loose set of administrative policies and much fewer dedicated resources in comparison with their mainframe counterparts. In fact, advanced storage techniques, such as hierarchical storage management (HSM), are still in their infancy in Windows NT shops. However, as storage needs grow, Windows NT shops will need the same scalable and robust solutions their mainframe counterparts employ now.

Backup Confusion

One of the biggest problems in the Windows NT marketplace is that it is tough to find a complete backup solution from any one vendor. Jerry Hoetger, group product marketing manager, NetWorker business unit, Legato Systems Inc. (Palo Alto, Calif., www.legato.com), says, "each vendor brings totally disparate technologies to the table with little leveraging of each other's strengths."

A variety of companies are working to solving this problem both individually and through partnerships, but the emergence of new technologies -- such as SANs -- to Windows NT, as well as to legacy environments, is making this goal a moving target.

Despite the vast differences between the two environments, backup vendors believe that tapping into the long tradition of rigid storage management and backup techniques found in the mainframe world holds the key to meeting the future demands of Windows NT shops.

The biggest challenges to achieving "mainframe-class backup" center around the management issues inherent in a distributed environment and the performance issues resulting from Windows NT's relative infancy and lack of robustness when compared with legacy systems. Centralized backup must evolve into more than just giving an operator the ability to view a bunch of backup servers on one screen.

"The distributed nature of Windows NT deployment necessitates single point of control and single point of administration for these environments," says Legato's Hoetger. "Administrators want to retain centralized administration while decentralizing the location of the tape devices and libraries that house the backups, archives and migrated data."

Most popular NT backup packages provide a centralized console for remote backup administration. However, centralized management needs to go a step further. For example, storing master configurations and policies on a master server greatly simplifies system management and reduces the resources needed to oversee backup operations. Centralized management coupled with more use of automated tape management devices can significantly reduce the amount of human intervention needed to keep backups running across the entire enterprise. NT backup vendors are not yet doing this to the extent that mainframe backup vendors have, such as with master/slave policy distribution and centralized library management.

John McIntosh, analyst with Boulder, Colo.-based analyst firm Storage Systems Marketing Inc., says that "NT’s ability to support multiple I/O threads is limited and represents a fundamental architecture problem. In addition, NT has no true distributed lock capability that is necessary for many data applications, such as storage-area networks."

While Microsoft Corp. continues to make Windows NT more robust in these areas, Seagate Software and other backup vendors continue to develop technology such as software agents to help get backups done faster. Agents residing on target systems compress and push data to the backup server rather than force the backup server to do all the work. Techniques such as replication push data to a central server, where it can be backed up to a local tape drive rather than entirely over the network. Additionally, applying RAID techniques to tape backup can go a long way in reducing the amount of time it takes to get data backed up during scheduled backup windows.

Storage-area Networks

SANs isolate internodal communication using clustering techniques over a high-speed medium such as Fibre Channel while users access applications over the standard LAN. While SANs are not restricted to backup applications, backup vendors are already taking advantage of the high-speed technology. Hewlett-Packard Co. recently introduced tape devices connected to a SAN via a SCSI-to-Fibre-Channel bridge. HP's current implementation will allow the attachment of two SCSI tape drives to a SAN, but future implementations will increase that number.

Technology such as SANs will allow managers to apply the age-old techniques of multiplexing and parallelism to their backup strategies without having to tax their user network. Mark Wanger, strategic planner at the systems storage division of HP, explains that while the idea of installing a separate "backup subnet" has been employed rather inexpensively for some time now, the use of clustering and Fibre Channel technologies will take backup to a new level.

Legato's Hoetger points out other advantages of applying parallelism and multiplexing to backup: "Parallelism addresses the ability to back up multiple file systems at the same time. Multiplexing addresses the ability to back up data from multiple clients to the same tape drive, at the block level -- intermixing backup data from multiple clients, block by block. Only solutions that provide high levels of parallelism and multiplexing can exploit high-speed devices and maximize data throughput rates for storage management."

According to Legato, the company pioneered the idea of parallelism with its NetWorker application. NetWorker 5.1, for example, takes advantage of all the above-mentioned techniques and supports Windows NT.

Hoetger points out that once backup performs at this level, "backup windows" will disappear, allowing backup to go on in the background on a continuous basis. Veritas' Ballard even predicts that "backup will become just another Windows NT service, operating transparently to end users."

Hierarchical Storage Management

At the moment, legacy systems have it all over Windows NT shops when it comes to managing data in a centralized and highly organized manner. One technique used in mainframe shops that has been very slow to catch on in the NT space is HSM, a tiered storage environment that stages data from disk off to slower, less expensive media on the basis of the frequency of use or other user-defined parameters. At the bottom rung of the HSM ladder are tape libraries, but these systems provide archive and retrieval – but don’t function in a backup capacity.

HSM would never preclude the need for robust backup systems. According to HP's Wanger, "I think of HSM as a way to decrease management effort and cost. If a file is migrated to another device, it is still the primary copy and therefore is unprotected! HSM just moves the data and makes it imperative that the backup system track movement. "

Veritas' Ballard agrees: "The biggest problem with relying on HSM for data recovery is that the files are moved, not copied. If the user deletes the stub or index file off of the primary storage disk, access to the whole file is lost. With backups, the file is copied out to a different media and is restored in whole when requested. HSM moves the file off the primary disk and leaves the stub/pointer file. HSM is for disk management/cleaning, not backup."

While HSM will become an important technology for Windows NT shops in the near future, reliable, robust Windows NT backup will occupy center stage for a while.

For the NT Administrator

Here are some key things for NT administrators to observe about NT backup.

  • There’s more to NT backup than copying files to tape. Many NT administrators look no farther than the speed of their tape drives to judge backup performance. This is a mistake.
  • There are more players in town than Computer Associates and Seagate. Expect to soon consider more vendors on your backup short list.
  • NT backup will have to be more robust if administrators expect NT to push its way further into the data center. NT backup solutions will have to reliably handle a larger volume of distributed data. Centralized management must go farther than just providing a console to see all backup servers on the network.
  • NT administrators will have to spend more time learning about disk and data management techniques. HSM will become a reality in the NT environment. Backup will be a critical component of total storage management. Right now, this is an afterthought in the NT world. Protecting existing data is just as important and requires just as much technology as processing current data.
  • The traditional network stifles backup performance. Clustering and Fibre Channel technologies can help. Although typically not used just for enhancing backup performance, backup applications and hardware should take advantage of these new technologies.

As Windows NT pushes its way further into the enterprise data center, reliable, robust backup will be absolutely a mission-critical necessity. There's much to be gained by analyzing the success of mainframe backup solutions. Windows NT administrators do well to learn a few lessons from the legacy masters in this area. --David B. Miller is a lead consultant for Integrated Systems Consulting Group Inc. (Wayne, Pa.). Contact him at dbm560810@aol.com.