Comparing disk-to-disk (to tape) backup models

A number of options exist to overcome some of the inefficiencies of traditional tape-based backup approaches.

By Marcia Reid Martin

The trouble with backup is that it hasn't kept pace with other growth and expansion trends in information technology.

  • Backup has gotten faster, but the growth of data to be protected has outstripped it. The result is that backup jobs take longer.
  • The backup window is too short, and the "risk window" is too long. The need for continuous application availability requires spending less time making backups. But the increasing value and change rate of data dictates making backups more often. Two critical forces are pushing backup in opposite directions.
  • Recovery takes too long. The business processes that can't be shut down for two hours each day to make backups are devastated by the 16 hours typically required for complete system recovery.

Today, the industry is searching for ways to make much-needed improvements in backup-and-recovery systems. Fortunately, recent breakthroughs in disk technology, enabling very low-cost, high-capacity disk arrays, are making it practical to enhance the backup process by adding secondary disk to the backup hierarchy. This article examines several ways to do this and discusses the relative merits of each.

Can disk save the day?

Disk subsystems are doing a good job of keeping pace with the rates of growth and change in information systems. Disk drive prices are falling, capacities are increasing, and reliability is high. As a result, users are increasingly considering the use of extra disk systems to improve the backup process. Actually, disk-based backup has existed for several years, but high cost has limited its use to only the most critical business applications. Only in the last year has the emergence of low-cost, high-capacity ATA arrays made disk-enhanced data protection a practical option for most companies.

To evaluate the emerging solutions, it's necessary to first determine what value can be gained from adding secondary disk subsystems to the backup architecture. What will make backup work better, less disruptively, and more reliably? The table above suggests some critical value metrics to apply to this analysis.

Click here to enlarge image


These metrics will help measure the effectiveness of the different disk-based backup solutions. The options available today can be divided into two broad categories.

Target substitution (disk as tape)

At the high end of the disk array market, replication-based technologies have been used to improve backup processes for several years. However, early attempts at introducing disk-based data protection into the midrange have not focused on replication, but on replacing tape as the primary target for backup data streams. Why? First, because it's relatively easy. Second, because replacing tape doesn't impact the primary storage or the primary server.

Today's controller-based replication approaches may require replacing the primary storage and are costly. And software-based replication can place a heavy computational burden on the application server. These two negatives probably explain the early emphasis in the midrange of tape-replacement solutions for backup enhancement. But is tape replacement the best approach?

Today, there are two common models for tape replacement: virtual libraries and disk-to-disk backup. In the former, a disk subsystem is given some basic emulation intelligence that permits it to appear as a tape subsystem. The backup application remains unchanged and unaware that the backup "tapes" it creates are actually stored on disk.

On the other hand, in disk-to-disk backup, the backup application is aware of the change. The backup application creates "backup sets" in a directory on a file-structured disk array, rather than on tape. These backup sets are cataloged and organized just as tapes would be. A special recovery program extracts the data, just as with tape.

Both solutions are similar in the benefits they provide, which is shown in the table below.

Click here to enlarge image


However, these approaches have limited benefits. The main benefits include much faster recovery of single files, and less operator intervention than is required with even automated tape libraries. Backup administration is simplified, because no matter how many backup jobs are aimed at a single disk target, there's always room for more. And backup operations fail far less frequently. In addition, RAID 3 or RAID 5 can be used to achieve very high reliability.

Click here to enlarge image


In contrast, disaster recovery and database recovery benefit little from these architectures. Tape systems are usually faster than disk drives for bulk data transfer, so performing a full-volume restore via disk-to-disk can take longer than restoring the same volume from tape.

Disk-aware D2D architectures

New backup applications are beginning to emerge that take more advantage of the media characteristics that differentiate disk from tape. With these approaches, a disk backup volume is able to be online all the time, and it is randomly (as opposed to sequentially) accessible, so that data can be modified without reorganizing the entire volume. Some of the improvements over traditional backup using a disk target include

  • Frequent or continuous capture of changes—The backup agent (running where the primary data is stored) notices changes when they happen and copies them (either immediately or in scheduled batches) to the backup disk, shortening the window of risk;
  • Synthetic full backups—A full backup set maintained on disk can be updated directly from the incoming partial (incremental or differential) backups, so that it always represents a fairly recent point in time. This shortens recovery time;
  • Finer granularity—If a data object changes several times between backups, all the versions can potentially be captured.

The table above shows how these advantages improve the backup value proposition.

Impediments to adoption of new products built on this model are mostly non-technical in nature:

  • Enterprises are heavily invested in established backup applications;
  • Adopting a different backup model requires restructuring many policies and procedures;
  • Most of the early proponents of this approach are start-ups, and some companies have policies against acquiring critical technology from small firms; and
  • Backup applications are complex. The established vendors have spent many years bringing their products to their present level of stability.

There is one technical issue that disk-centric backup vendors must address, however, and ironically it concerns tape.


While there are many benefits to storing backup data online (on disk), there are some essential backup requirements that an online disk array doesn't meet. For example, the disk media isn't removable for off-site storage. A low-cost disk array may be large enough to hold weeks of backups, but it will not hold volumes earmarked for long-term retention. In any case, online storage isn't secure enough for long-term copies. For this reason, it is important to provide the ability to eventually move backup sets off to tape or to other removable media.

Migrating backup sets to tape is easy for virtual-tape systems and just as easy for backup applications that create monolithic backup sets on disk. Disk-aware backup applications that adapt the structure of the backup data to disk face the additional challenge of providing a mechanism for storing selected backup images on tape as well.

Continuous backup vs. data replication

In the March 2003 issue of InfoStor (see "Comparing replication techniques: Snapshots and mirroring," p. 38), we looked at how point-in-time replication technologies can be used to enhance backup. In this article, we explored emerging approaches to disk-based backup. To some extent, a continuous-capture backup strategy can reduce the need to use point-in-time technology to "freeze" application images for backup. Interestingly, a new class of replication technologies is emerging that has many features in common with disk-to-disk backup. The next article in this series will discuss journaling data replication techniques.

Marcia Reid Martin is a consulting software engineer at StorageTek (www.storagetek.com) in Louisville, CO.

This article was originally published on June 01, 2003