Introduction
Tremendous progress has been made in tape storage technology since the announcement of IBM's first tape drive—the IBM 726 tape unit—in 1952. The IBM 726, depicted in Figure 1, was able to store data at 6.1 kB/s at an areal density of 1400 bit/in2 and had a capacity of approximately 2.3 MB.
In 2011, IBM began shipping the TS1140 tape drive with an uncompressed capacity of 4 TB in a single tape cartridge and a native sustained data transfer rate of 250 MB/s, depicted in Figure 2. This translates into an improvement of 10,000× in data rate and more than 1,000,000× in areal density (i.e. more than 6 orders of magnitude). IBM's latest tape library, the TS3500, is depicted in Figure 3 and can store up to 900 PB of uncompressed TS1140 data.
Currently, tape systems play a central role in archival and disaster recovery applications and are the technology of choice for tertiary bulk storage in enterprise systems due to their very low total cost of ownership, low power consumption, high volumetric density and very high reliability, see Figure 4. Tape systems also become an integral part of active archives, in which data can be migrated to the most appropriate storage tier (e.g. SSD, HDD, tape) and where users can access data from all storage tiers through a common filesystem that represents all the tiers as a single namespace.
In order to maintain their dominance in these applications, tape-based storage systems need to match or exceed the performance growth of hard-disk systems, while maintaining or improving their price/GByte advantage. Fortunately, there is still significant potential to continue scaling of tape technology.
For example, in 2010 in collaboration with FUJIFILM, IBM demonstrated the potential for recording at an areal density of 29.5 Gb/in2 using a prototype BaFe tape fabricated using low cost particulate coating technology [1]. According to the 2012 Information Storage Industry Consortium (INSIC) tape road map, such an areal density will enable a cartridge capacity of 64 terabytes in the 2020 timeframe. Moreover, a recent feasibility study indicates that there is further potential to continue scaling tape towards areal densities of 100 Gb/in2 [2].
This work clearly indicates that there is enormous potential to continue scaling tape systems for many years to come. However, maintaining both the growth in the areal density and data rate of tape systems while maintaining other desirable attributes such as backward compatibility, poses complex technical challenges that span diverse disciplines including signal processing and servo control, materials science, magnetic recording physics and mechatronics.
In order to increase tape cartridge capacity while maintaining the current cartridge form factor, either the area of tape that is available to store data must be increased and/or the areal density must be increased. One approach to increase the area available for storing data is to improve the format efficiency, e.g. through the use of more efficient coding schemes.
An even simpler approach is to increase the available surface area of the tape, however, there are some limitations. For example, there is a strong incentive to preserve the half inch tape width used in linear tape drives due to the large investment in equipment for manufacturing tape in this format. Similarly, it is desirable to preserve the cartridge form factor due to the large installed base of automated tape libraries that are compatible with the current formats.
- Figure 1. IBM 726 Tape Unit from 1952 with a reel capacity of 2.3 MB.
- Figure 2. TS1140 Tape Drive from 2011 with a native cartridge capacity of 4 TB.
- Figure 3. TS3500 tape library, scalable to over 300,000 LTO cartridges or more that 900 PB of uncompressed TS1140 data.
- Figure 4. Smart tiering.
On the other hand, the tape length can be increased without changing the cartridge form factor if the tape thickness is simultaneously decreased. Unfortunately, this renders the tape more fragile and necessitates advances in the tape transport system and tape path design.
Over the next 10 years the INSIC Tape roadmap predicts a 32× increase in cartridge capacity, of which tape length scaling is expected to contribute a roughly 50% increase in capacity (i.e. 1.5×) with an additional 15% capacity increase (1.15×) due to improvements in format efficiency. Hence, the majority of the expected capacity increases will have to be achieved by scaling the linear and track density.
While continued incremental increases in linear density are expected in the future, there is a much larger potential for increasing track density by reducing the width of the individual data tracks. In order to exploit this potential one has to minimize tape lateral motion and compensate for any remaining disturbances by adjusting the position of the write/read heads dynamically to follow the disturbances.
For very high track densities, positioning control down to the nanometer scale will be required. Such precise control necessitates significant improvements in the performance of all elements of the track follow servo system, including the servo pattern, the servo channel, the head actuator and the track-follow servo controller [2]-[6]. In addition, continued improvements in tape media dimensional stability (TDS), e.g with respect to changes in environmental conditions, or the implementation of a TDS control scheme will be required to maintain all of the write/read elements on track under the varying environmental conditions in which tape drives operate.
A third challenge associated with track density scaling arises from the need to reduce the width of the data read element in proportion to the width of the data track. This results in a reduction in signal to noise ratio (SNR) of the read-back signal that must be compensated for by a combination of improved media technology, improved read transducers and improvements in the read channel [7].
Another significant scaling challenge results from track edge effects such as side erasing and track edge curvature. As track widths are reduced these effects become large relative to the track width and must be minimized through innovative design of the write transducers.
In the area of media technology, the SNR performance of particulate media can be improved by reducing the volume of the particles, which reduces media noise. However, this has the consequence of reducing the thermal stability of the particles, which must be compensated for by increasing the anisotropy energy density of the particles. This in turn requires the development of write transducers capable of producing sufficiently large magnetic fields to write these particles.
A second approach is to make the media smoother in order to reduce the magnetic spacing, i.e. the distance between the read/write transducers and the magnetic coating. However, increasing the tape smoothness tends to increase tape-head friction, which in turn reduces the “runability” of the media and leads to high frequency velocity variations that can degrade the performance of the data detection process.
Continued scaling of the data rate of tape systems also poses significant technical challenges. Historically, the data rates of tape systems have been increased through a combination of enhanced linear density and higher tape speed, and by increasing the number of write/read transducers that operate in parallel.
IBM’s latest enterprise drive, the TS1140, writes and reads 32 data tracks in parallel. In the future, linear density increases are expected to be more modest than in the past. There are also limits to the maximum tape speed that can be achieved without increasing the disturbances to the track follow system to unacceptable levels. Hence the number of parallel channels will likely be scaled more aggressively in the future.
However, increasing the number of channels while maintaining a constant head span to minimize TDS effects leads to a reduced spacing between transducers. This is particularly challenging for write transducers due to fabrication issues and due to the possibility of cross-talk between adjacent writers at small transducer pitch. Avoiding such cross-talk effects requires careful design of the write transducers. In addition, adding more transducers in the head results in an increase in the number of electrical connections that must be routed over the long flex cable connecting the head to the electronics card of the drive.
Increasing the data rate of an individual channel by increasing linear density and/or tape speed is also quite challenging, due to the impedance of the flex cable. Some of these challenges can be addressed by moving some of the front-end electronics off the drive card and directly adjacent to the head on the flex cable, using flip-chip technology.
The ultimate key to maintaining and increasing the success of tape systems in the market is to improve the usability of tape. In the past, one of the obstacles to the more widespread use of tape technology has been the difficulty of using tape in a general or stand-alone context. Hard disks provide random access to data and generally contain a file index managed by a file system. These files can be accessed using standard sets of application programming interfaces (APIs) via various operating systems and applications.
Tape, in contrast, is written in a linear sequential fashion using a technique called “shingling” which provides backward write functionality, but also imposes the restriction that new data can only be appended and that previously written areas can only be reclaimed if the entire cartridge is reclaimed. In traditional tape systems, an index of the files recorded on a given cartridge is typically only kept in an external database managed by an application such as a proprietary back-up application. The requirement to access an external database to retrieve data renders data on tape much less portable and accessible than with alternative storage methods, such as a HDD or a USB drive.
To address these deficiencies, a new long-term file system (LTFS) has recently been introduced into LTO tape-drive systems starting with LTO-5 and into IBM’s enterprise tape drive systems to enable efficient access to tape using standard system tools and interfaces. LTFS is implemented using the dual-partition capabilities introduced in the LTO-5 format and also supported in recent enterprise products. A so-called index partition is used for writing the index, and the second, much larger partition for the data itself. With this new file system, files and directories show up on the desktop with a directory listing. Users can “drag and drop” files to and from tape and can run applications developed for disk systems.
All these features help to reduce the costs associated with using tape and eliminate the dependency on a middleware layer. In addition, tape becomes cross-platform-portable, enabling and facilitating the sharing of data between platforms. These features enable significant new use cases for tape such as video archives, storage of medical images, etc.
In the future, we envision tape systems becoming even easier to use and manage and more closely integrated in a tiered storage hierarchy. In such a system, data will be moved seamlessly between different storage tiers automatically, depending on policies and access patterns that are “learned” by the system, in order to minimize cost and maximize performance.
Our research in tape storage systems aims to address all of these challenges and to develop the technologies which will be essential components of future tape systems. Specifically, we currently focus on:
- devising novel signal processing and coding techniques,
- optimizing the track follow servo system, including:
- servo format design,
- servo channel design,
- actuator design,
- track follow servo control
- use of MEMS accelerometers for enhanced vibration rejection
- optimizing the tape transport design and reel-to-reel control system
- design of advanced write heads
- planar servo writer technology
- integrated electronics
- characterization of prototype media and advanced read head technologies
- integrating tape into the storage hierarchy
References
[1] 29.5 Gb/in2 Recording Areal Density on Barium Ferrite Tape, G. Cherubini, R.D. Cideciyan, L. Dellmann, E. Eleftheriou, W. Haeberle, J. Jelitto, V. Kartik, M. Lantz, S. Oelcer, A. Pantazi, H. Rothuizen, D. Berman, W. Imaino, P.-O. Jubert, G. McClelland, P. Koeppe, K. Tsuruta, T. Harasawa, Y. Murata, A. Musha, H. Noguchi, H. Ohtsu, O. Shimizu, and R. Suzuki, IEEE Transactions on Magnetics 47(1) (January 2011).
[2] Scaling Tape-Recording Areal Densities to 100 Gbit/in2, A. Argumedo, D. Berman, R. Biskeborn, G. Cherubini, R. Cideciyan, E. Eleftheriou, W. Häberle, D. Hellman, W. Imaino, J. Jelitto, K. Judd, P.-O. Jubert, M.A. Lantz, G.M. McClelland, T. Mittelholzer,S. Narayan, S. Ölçer, IBM J. Res. Develop. (Storage Technologies and Systems) 52(4/5) (2008).
[3] Characterization of Timing Based Servo Signals, G. Cherubini, R.D. Cideciyan, E. Eleftheriou, P.V. Koeppe, Digest of Technical Papers IEEE Int'l Magnetics Conf. "INTERMAG 2008," Madrid, Spain, May 5-8, 2008, pp. 600-601.
[4] Synchronous Servo Channel Design for Tape Drive Systems, G. Cherubini, E. Eleftheriou, J. Jelitto, R. Hutchins, Proc. 17th Annual ASME Information Storage and Processing Systems Conf. "ISPS 2007," Santa Clara, CA, June 2007, pp. 160-162.
[5] Nanoscale Track-follow Performance for Flexible Tape Media, M. A. Lantz, A. Pantazi, G. Cherubini and J. Jelitto, Proceedings of the 18th IFAC World Congress, Milano (Italy) August 28 - September 2, 2011.
[6] Servo-Pattern Design and Track-Following Control for Nanometer Head Positioning on Flexible Tape Media, M. Lantz, G. Cherubini, A. Pantazi, J. Jelitto, IEEE Transactions on Control Systems Technology, Volume: 20, Issue: 2, March 2012, pp. 369-381.
[7] Adaptive Noise-Predictive Maximum-Likelihood (NPML) Data Detection for Magnetic Tape Storage Systems, E. Eleftheriou, S. Ölçer, R.A. Hutchins, IBM J. Res. Develop. 54(2) (2010) Paper 7.