Disk-to-disk-to-tape backup alternatives, part 2

D2D2T devices can be categorized as appliances, disk libraries, or virtual tape systems, all of which are viable solutions.

By Phil Goodwin

In part 1 of our series examining disk-to-disk-to-tape (D2D2T) solutions, we were primarily concerned with establishing the difference between data replication technologies and D2D2T (see InfoStor, October 2004, p. 37). Upon full examination of the issues, it becomes clear these two technologies are "and" solutions rather than "or" ones. Specifically, replication solves the data-protection problem. Replication technologies include snapshots (logical metadata copies), mirrors (physical copies), and remote replication (physical copies at a second site). When deployed in combination, these technologies establish a continuum that protects the organization from nearly any cause of data loss from inadvertent deletion to site disaster. However, even when replication technologies are deployed, tape backup remains a best practice operation for both low-cost, long-term archiving as well as a physical break in the data flow for "last chance" data recovery.

Although backup and recovery (B/R) contributes to data protection, it also introduces numerous problems into storage operations. Tape products, being mechanical devices, have higher failure rates than purely electronic components. Tape devices tend to overheat when proper cleaning and maintenance schedules are not maintained and tend to have higher error rates due to dirty tape heads. The tape media itself is subject to degradation due to time, environmental damage (e.g., heat and humidity), breakage, and surface wear. Moreover, B/R jobs may fail due to server downtime (both planned and unplanned), network outages or constrictions, software errors, and failure to complete within the backup window. In fact, we estimate that 5% to 20% of nightly tape-based B/R jobs fail on average.

When B/R jobs fail, the IT organization is compelled to take action because the failure places the organization at risk of data loss. Therefore, storage administrators must examine failed jobs on a daily basis, determine the root cause, decide if remedial action is necessary, and then implement that action. Of course, this is in addition to the daily "care-and-feeding" tasks of moving tapes off-site/on-site, replenishing scratch pools, and so on. Tape management is such a manual-intensive operation that we estimate that it can consume 60% to 70% of the effort associated with storage management.

D2D2T solves the B/R problem better than any solution developed to date. Of course, it does not improve the mechanical properties of either disk devices or tape media, nor does it solve any of the daily care- and-feeding issues. Functionally, D2D2T is a disk device to which a backup job first moves data and then later spools the data off to tape. D2D2T does, however, solve many of the other key issues:

  • Backup window: Benchmarks have shown that disk-to-disk replication can be up to 10 times faster than tape backups.
  • Mechanical limitations: Backing up to a D2D2T device eliminates the problems associated with tape drive and tape media failures. These devices may fail just as often, but because the disk backup contains a copy of the data, the organization is protected from the possibility of data loss.
  • Manual effort: The biggest savings is, perhaps, in regard to the effort associated with recovering from failed jobs. With the company's risk mitigated by a backup copy residing on disk, the urgency to restart a failed job is greatly reduced. In fact, the administrator can most likely focus on back-to-back failures (i.e., jobs that have failed two nights in a row). Only the most critical jobs that must be rotated off-site on a daily basis require immediate attention. As a result, the storage administrator's tape-related duties may be reduced by as much as one or two hours per day.

With these benefits in mind, the next question is, "What is the difference between the various D2D2T products?" Three broad categories of products have emerged to address D2D2T: disk-to-disk (nearline) appliances, disk libraries, and virtual tape solutions. The succeeding sections will examine each of these categories and representative products in each category.


The appliance category is rapidly becoming crowded with a variety of entrants. Three examples of vendors with D2D or D2D2T appliances are Data Domain, Network Appliance, and Nexsan Technologies.

Data Domain is an emerging vendor targeting the D2D2T market. Its principal product is the DD200, which the company describes as a "disk recovery appliance."

The DD200 is a 4U package capable of holding 23TB of usable storage utilizing ATA disks. Data is moved to the DD200 using standard, off-the-shelf backup software. This is an important aspect, because it minimizes changes to operational backup procedures (and applies to other vendors as well). The key differentiation for the DD200 is Data Domain's claim of a 20x average compression rate combined with full backup data verification. This compression rate is achieved through a process of moving sub-block-level incremental changes to disk.

The DD200 has other important functionality, including the ability to provide asynchronous replication between DD200 units at different locations. Data Domain uses its compression technology to minimize bandwidth and claims that a 96% reduction in bandwidth needed for weekly replication can be achieved. In addition, the company suggests that the DD200 can be used for "mid-term" archival storage for up to six months.

Network Appliance can rightly be credited as the first large vendor to recognize and popularize the D2D2T market with its NearStore R100 in early 2002. Currently, the company is on its third-generation product, the R200, and claims to have shipped more than 30 petabytes of NearStore storage.

NearStore is based on ATA disk technology (unlike NetApp's other products, which are based on Fibre Channel drives), but shares the same DataONTAP operating system and WAFL file system as all NetApp products. NearStore can use either standard backup software to move data onto NearStore and then to tape, or NDMP.

NearStore has two key differentiators. First, Network Appliance offers the product with RAID DP (double parity), which is similar to RAID 6 in that it has two parity drives for each 14 disks. However, the company claims to have solved the performance deficiencies of RAID 6. NetApp points out that, as the number of disk devices increase, the chances of a double disk failure in a RAID group mathematically increase to a probability. Moreover, recovering a single high-density SATA drive (e.g., 320GB) can take 18 hours or more. (As a side note, one of Diogenes' clients recently suffered such a double-disk failure in an ATA RAID group, and it took the company fully three weeks to recover.)

Second, NearStore can eliminate or reduce the need to back up remote systems to tape. Remote system backup becomes problematic if the local staff does not have sufficient IT skills to effectively manage the tape environment. NetApp offers the ability to replicate changed blocks from the remote site using SnapVault to move the data from the remote site to a central NearStore system. Then, the data can be moved to tape by the data-center staff.

Along with Network Appliance, Nexsan Technologies is one of the older entrants in the D2D market. Its ATA-series products range from the ATAbaby up to the 42-drive, 4U ATAbeast. As the names imply, these systems are ATA- and SATA-based, but will support SAS drives in the future as well. Last month, Nexsan introduced the 1U, 3.2TB SATAblade subsystem, which can be used for D2D backup or as primary storage (for coverage of the SATAblade, see the New Products section, p. 40).

Nexsan's key differentiation is its "Maytag" approach to engineering. Although washing machines look the same on the outside, Maytag emphasizes what's inside. Similarly, Nexsan points to what it believes is superior cooling, vibration isolation, and failure isolation that lead to fewer soft errors, better performance, and longer disk life. Nexsan also highlights what it calls "extreme density." Whether this extra engineering results in "cleaner clothes" may be debatable, but organizations looking for more ruggedized systems or an edge in RAS (reliability, availability, and serviceability) or footprint will take a serious look at Nexsan.

Functionally, the ATA-series systems offer configuration flexibility to behave like a tape library or a backup appliance using either qualified third-party backup software or replication software. Moreover, a subsystem can be partitioned into different functional systems, such that it is simultaneously an appliance, tape library, and directly accessible disk subsystem. Nexsan offers the option to store data in "full file format," which allows a partition to be addressed as a mount point for direct processing.

Disk libraries

EMC can be credited with creating the "disk library" category of D2D2T products when it introduced the CLARiiON Disk Library (CDL) earlier this year. CDL is based on the CLARiiON CX300 and CX700 systems using ATA drives. The key differentiator between disk libraries and appliances is that disk libraries emulate both the robotic device as well as the tape device. In other words, the disk library "answers up" to the B/R software (using SCSI commands) as a specific tape library with specific tape devices. As a result, the backup software behaves exactly as it would with a tape library. Operationally, the backup process is the same as always, with the exception that tape cartridges (obviously) never need to be ejected. However, the storage administrator must still be aware that the B/R software will "think" that it is limited to the number of slots and drives that a specific library would support. Similar to D2D2T appliances, the CLARiiON Disk Library uses off-the-shelf B/R software to physically move the data. Currently, CDL emulates the Quantum ATL 7100, ADIC I2000, IBM 3584, and StorageTek tape libraries. Drives emulated include LTO, DLT/SuperDLT, and STK 98XX/99XX. Data on the CDL can be migrated to a physical tape library, and in fact, carries the advantage that the data is already properly blocked for tape write operations (i.e., fixed block sizes rather than variable block sizes) for optimum data movement.

Virtual tape

Although some vendors in the D2D2T market have adopted the moniker "virtual tape" for their products, we take a more purist view toward the definition of virtual tape. Virtual tape systems, such as IBM's VTS and StorageTek's VSM, were originally designed for the mainframe market. Mainframe systems notoriously use tape very inefficiently, often putting only a few megabytes of data on a 10GB or 20GB tape cartridge.

By using virtual tape on the front-end of physical tape, virtual tape is able to densely pack and optimize physical tape, often saving tens or hundreds of thousands of dollars in media costs.

Both VTS and VSM use disk to emulate tape drives, much as a disk library does. However, the data is moved using the mainframe's own sequential access methods rather than B/R software. Virtual tape may also utilize hierarchical storage management (HSM) software to move the data from virtual drives to physical drives. Thus, traditional virtual tape really addresses mainframe data access issues, and neither data protection nor B/R.

Overland Storage positions its REO 1000/4000/9000 systems as "backup appliances using virtual tape." We are more inclined to categorize these products as disk libraries, although they do not truly emulate a library as does EMC's CDL. (It is not really necessary to emulate a robotic device, because there is no tape movement and these commands can be ignored.) Regardless of moniker, however, the REO products are designed to solve the B/R problem. These products emulate tape devices and allow B/R software to write to the disk as though it were a sequential device. As with some other D2D2T products, administrators can either move the data to tape from the REO, or simply have two data streams-one to the emulated tape drive and one to an actual tape drive.

The main differentiator for REO systems is Overland's ProtectionPAC software. This software provides a "dynamic virtual tape" that matches the size of the virtual drive to the backup job. As a result, there is no wasted space in pre-allocated drive sizes, nor do administrators need to worry about the size of a virtual library. Moreover, the "virtual media" is managed just as physical media would, including catalog management, scratch pools, retention, and expiration.

In many respects, the various products in the D2D2T category have more commonalities than differences. All of them use commercial B/R software and maintain current operational procedures with minimal change. None of them replace B/R, but do enhance both operational efficiency as well as data-recovery speed. So, does it really matter if a device emulates tape or not? We tend to think not, and that each of these methodologies is valid. Other factors will likely drive the purchase decision. In some cases this might be a specific feature offered by one product, or it might be vendor preference, packaging, or price. In any event, don't allow confusing "marketecture" to eliminate one solution before the evaluation even begins.

Phil Goodwin is president of Diogenes Analytical Laboratories (www.diogeneslab.com), an IT consulting and product evaluation firm, in Boulder, CO.

For more information on disk-based backup, see the following articles that have appeared in InfoStor:
"Disk-to-disk-to-tape vs. replication, part 1," October 2004, p. 37
"Users shift to disk-based backup," October 2004, p. 40
"Emerging and innovative disaster-recovery strategies," August 2004, p. 26.
"Backup is dead. Long live backup!", August 2004, p. 28
"Is online disk backup a fit for SMBs?", July 2004, p. 1
"Replication options multiply," June 2004, p. 1

This article was originally published on November 01, 2004