The technology of tape automation.

This year, 2002, is the 40th anniversary of the automated tape library (ATL). IBM delivered the Tractor system to the National Security Agency early in 1962. This massive system which contained six tape drives and 360 15-pound tape cartridges was used continuously for 14 years. It is a measure of the vitality of this technology that the capacity of the Tractor system (43GB), which was unheard of at the time, is a fraction of the capacity of a single cartridge in a state-of-the-art automated tape library today.

Automated tape libraries have played an increasingly important role in data protection solutions over the intervening four decades. While there were a number of special-purpose systems introduced in the early years, it was a series of key developments beginning in the mid-1980s that have led to today's ATL market which IDC estimates will reach 100,000 ATL shipments per year by 2003. IDC forecasts an additional shipment of more than 160,000 single-drive "autoloaders" per year by 2003.

The first critical technology was digital cartridge tape, introduced by IBM as the 3480 tape system, in 1984. Automation systems to support the 3480 tape system became available very shortly later. The 3480 provided, for the first time, a standard tape system that was amenable to automation; a system that has survived in a highly-evolved form to the present day.

Several years after the 3480, Exabyte introduced a very high-density tape system based upon the commercial 8mm tape cassette. This system adapted helical recording techniques that had been developed for video recording to provide digital recording at very high densities when contrasted with the linear recording techniques that had traditionally been used for digital recording. Once again, the cartridge tape format permitted the introduction of automation systems to support the 8mm cassette format soon after the initial introduction of the format itself. Once again, this format has proven durable. Both the Exabyte system and the AIT 8mm system from Sony have technical roadmaps that suggest they can meet a range of needs for a considerable time to come.

One of the most significant developments in this area was the introduction of a high-density digital linear cartridge tape format, called DLT, by Digital Equipment Corporation in 1992. This format drew from the low-density DECtape technology, but provided density and cost advantages that rivaled helical tape systems while retaining the linear tape technology that was extensively used in the computer industry. DLT tape was sold to Quantum Corporation in 1994 and rapidly became the defacto standard tape system for the rapidly-growing network computing market. Automated tape libraries for DLT were introduced in 1994 by ATL Products, Inc.--now a part of Quantum Corp. At present, DLT-based systems represent more than one-half of the installed base of automated tape libraries and an even larger percentage of the growth of the installed base.

The most recent development has been the introduction in 2001 of two very similar high-performance tape systems: a new generation of DLT (called Super DLTtape--or SDLT) and a rival high-density linear cartridge tape system (called Linear Tape Open--or LTO) introduced by a consortium of industry partners including HP, IBM, and Seagate). SDLT and LTO tape cartridges are very similar and both have been supported by automation from their introductions.

While all ATLs share common technological challenges, the dominant issues for ATL design and implementation have evolved significantly over time. We will briefly discuss below some of the dominant issues over the past 15 years.

ATLs represented highly complex electro-mechanical systems in an industry that, on the whole, was moving rapidly to purely electronic systems. The function of transporting, importing, and exporting data cartridges was one of the last physical activities left in the data center.

Since a key element in the economic justification for tape automation was the elimination of expensive manual intervention, reliable autonomous operation of ATLs has been a critical requirement from the very beginning. At the same time, the complex kinematics required by the ATL function made it prohibitively expensive to use the traditional redundancy techniques that had been applied successfully to many other technologies. As a result, all components of ATLs had to be designed and tested for very high reliability. While the design techniques to achieve the required reliability were well-established, the testing required to gain an appropriate level of confidence in ATL reliability has remained a major challenge for ATL manufacturers since the electromechanical nature of these systems precludes many of the accelerated testing techniques that are available for electronic systems. At the present time, the large ATL installed base has provided a very high level of confidence that these systems have achieved a level of reliability that represents an effective economic balance for their applications.

ATLs typically represented among the highest value (and highest visibility) components in a computer system. As such they must be designed for very long life; while it is possible to replace small (e.g. 3.5 inch or "2U") rack mounted servers frequently, it is neither convenient nor supportive of CIO job security to replace rack-scale systems on a regular basis. Long life requirements have brought two different design issues to ATLs. The first, and most obvious, is that the systems must have extremely long "wear out" lives--typically 20 years or more. Fortunately, both the technology available for ATLs and their usage profiles make it possible to meet this requirement without a significant economic penalty. Since, in most environments, ATLs operate with relatively low duty cycles (typically 10% or less), it is practical to demonstrate this lifetime during system development and early deployment.

The second design issue resulting from long life, however, is less obvious. Over its useful life, an ATL must accommodate several generations of tape drives and tape cartridges. It has been typical, for example, for many DLT ATLs to have been through two or three generations of DLT technology and, in many cases these systems are in the process of being upgraded to SDLT. The ability to support this scale of field upgrade capability has had a major influence on system architecture over the past decade.

The 3480 was designed as a general-purpose tape system and therefore required high-speed automation to support cartridge exchange rates of hundreds of swaps per hour. Subsequent, higher density tape technologies have been optimized for backup applications and, therefore, much lower exchange rates--typically ten swaps per hour or less. While automation speed still has some advantages (e.g. minimizing automation duty cycle), the advantages are much smaller for these high-density tape technologies and modern ATLs have largely traded inherent automation speed for other critical characteristics such as cost, power, and physical size.

Over the long term, tape density (and, therefore, capacity) increases more slowly for any specific tape technology than it does for hard disk technology. At the same time, new tape technologies have typically introduced substantial step increases in capacity. For example, when DLT was introduced in the mid-1990s, the average ratio between a 3480 tape cartridge and a system disk was in the range of 2-to-1 while the same ratio for a DLT tape cartridge was over 20-to-1. In the nearly one decade that DLT has been in the market, this ratio for DLT (and LTO) technology had decreased to the 2-to-1 range. The effect of this dynamic is that the natural balance between the number of tape cartridges (i.e. capacity) and the number of tape drives (i.e. channels) tends to increase over time. For example, when DLT automation was introduced, typical 3480 ATLs contained thousands of cartridges while typical DLT ATLs contained hundreds of cartridges. This shifting balance brings two challenges to the design of ATLs; the first is the need to provide systems of increasing scale over the useful life of a tape technology to meet the needs of new users and the second is the need to provide economically viable expansion (scalability) capability to meet the needs of existing users. Over the past several years, scalability has become a critical differentiator among DLT ATL architectures.

This same dynamic, as well as the continuing reduction of the scale of all computer equipment, leads to an increasing requirement for physical density of ATLs over time. At the present time, for example, state-of-the art DLT and LTO ATLs have achieved a storage density of 120TB/[m.sup.3]--six times that of RAID storage arrays.

The final ATL technical design issue we will discuss here is driven by the need to share the ATL resources among as many users as possible to make effective utilization of these high value resources. The technical response to this requirement has been to migrate ATLs--particularly at the high end of the market--from direct-attach (i.e. SCSI) systems to network-attach (Fibre Channel or Gigabit Ethernet) systems. Native network interfaces (primarily Fibre Channel) have become the dominant interconnect for high performance ATLs and these interfaces are migrating down into the mainstream of the ATL market.

All of these issues, of course, have to be addressed in a market in which the economic value of accommodating a given unit of storage has declined extremely rapidly. In the mid-1980s, for example, ATLs could command a price of approximately $400/GB while today they are more typically in the range of $4/GB. These aggressive price reductions have come from a highly interactive combination of design improvements in the tape and cartridges themselves, the maturation of high-reliability automation system design and the system design to share large-scale ATL resources effectively among a number of independent applications.

In conclusion, automated tape libraries have represented a very dynamic market in which substantial technical evolution has been required to remain economically competitive. As we look to the future, many of these trends will continue. While we do not anticipate another disruptive technology change in the tape drive or cartridge technology over the next five years, we do see very aggressive technical roadmaps for all of the existing tape formats.

At the same time, the very aggressive development of disk technology will enable the integration of special-purpose disk arrays with ATLs to enhance backup performance and reliability. These integrated backup solutions will extend the ability of ATLs to address increasingly demanding backup requirements.

Fortunately, industry analysts such as IDC forecast a robust growth in the ATL market (typically in the region of 16% CAGR) which is sufficient to support the investment necessary to keep these products well-tuned to meet the backup and archiving requirements of the market.

www.quantumatl.com

Dr. Kevin C. Daly is the CTO of Quantum Storage Solutions Group (Irvine, CA).