The magnetic tape is one of those technologies which predates most modern computer users, and even some of the more established practitioners of the trade. We might assume that such a venerable item of technology may be running out of technological life, alas nothing could be further from the truth, with variants of established tape standards and new standards beginning to emerge.

In this month's feature we will stroll down memory lane and discuss the basic principles of established tape drive designs, and then explore some of the newer standards in the marketplace.

Magnetic Tape Technology

The magnetic tape drive started its existence as a development of magnetic wire recorders which came into use in the broadcast industry about a half century ago. The basic idea was very simple. By running the wire across a magnetic head containing a coil, an electric current flowing through the head would magnetise the wire, with the local strength of the magnetic field being roughly proportional to the amount of current flowing through the head at that time. By connecting a sensitive amplifier to the head, running the wire across the head induced a voltage in coil in the head, which could be used to recover the information, in this case voice or music, recorded on to the wire.

Wire was hardly an optimal medium for recording, and thus it did not take very long for the tape recorder to evolve, using a plastic tape coated with a thin layer of magnetic material. The reel to reel tape recorder very soon became the basis of what is termed the linear tape drive.

The most commonly used linear tape technology was the nine track 0.5" magnetic tape, which until the late eighties was the industry standard and can still be widely found in tape archives with decent histories.

The medium used in a nine track tape is a polyester film tape, which has a magnetic material not unlike that used in older generation disk platters applied to one side, and a carbon coating applied to the other to preclude the gathering of an electrostatic charge on the tape.

A typical drive head contained nine write coils and gaps, nine read coils and gaps, or nine read/write coils and gaps, and usually a single very wide erase coil and gap to wipe all nine tracks at once. Data was recorded by writing each of the eight bits in a byte at once, and a parity bit to the ninth winding. It was read back in like fashion, with logic circuits comparing the parity of the read byte against the read back parity bit. A servo controlled the speed of the tape so it was constant as it moved over the head.

Simple as this may be conceptually, there was considerable complexity in the bowels of the machine. The first area of interest was head and track skew, the source of many woes to support engineers during the seventies and eighties.

Head skew refers to a misalignment of the head relative to the tape. This has the interesting effect of making tapes unreadable on other drives, and requires that the head be carefully calibrated.

Track skew results from mechanical tolerances in the position of the coils and gaps within the head, causing some tracks to be written and read before or after others in the byte. This must be compensated by introducing electrical delays into the read and write channels, to ensure that all bits are written at the same time, and decoded at the same time. Track skew in any head would change over time as the surface of the head wore down with use. Having had the pleasure of deskewing many a nine track tape drive in my youth, I can assure the modern reader that the process of skew calibration made for a tedious half day at least.

The complexity of this generation of drives was not confined to head electronics alone, since high speed reading and writing meant that a decent loop of slack tape had to be available on either side of the head, since the huge inertia in the two tape spools meant they could never accelerate as fast as the capstan which drove the tape over the head. These loops of tape had to be tensioned, and with the exception of the very slow mechanically tensioned Cipher drives, it was usually done with a vacuum column. A loop of tape on either side of the capstan and head would sucked into a column by a vacuum pump. Most often, a series of holes was drilled into the column, behind which were positioned tiny vacuum actuated electrical switches. As the capstan wound the tape left to right, it would move up or down in the respective vacuum column, thereby tripping the switches. These switches would then activate the main spool motors to feed more tape from one reel into its column, and pull tape out of the other column to wind it on to its reel. The reel motors had massive coaxial clutch brakes to stop them quickly, as required, also tripped by the vacuum switches.

This generation of tape drives were electro-mechanical-pneumatic contraptions well and truly worthy of the C grade sixties and seventies sci-fi and spy thrillers of which they seem to be such a typical fixture.

Data was recorded typically using a Manchester or similar code, byte by byte, density determined by head design, magnetic material, and tape speed. Each tape record was preceded and followed by a preamble and postamble respectively, which were generated and detected by the drive electronics. Record sizes were frequently determined by the operating system, and the block size read and write parameter we can still find in tar and dd owes its origins to this period.

Needless to say the achievable capacity and write density was unspectacular, and very soon led to the next big leap in tape technology, which was the transition from linear recording to blocked recording.

A tape using blocked recording would operate with a fixed block size. Instead of arranging the bytes consecutively on the tape, the data was organised into a serial stream, to which a cyclic redundancy check polynomial or even forward error control data block was attached. This block was then rearranged in a buffer in the drive electronics, and split into nine chunks of equal length. When a tape was written, each of these nice chunks was serially into into one channel of the head.

Blocked recording meant that a tape with a given density could be packed with much more data, because most of all nine tracks could be used, and also because the error control allowed a much higher recording density on a given material.

The next big development in magtape technology was the Quarter Inch Cartridge or QIC tape, introduced in the seventies by 3M. The QIC-60 and later QIC-150 became a mainstay in the Unix workstation market and until the advent of the CD-ROM was the primary means of distributing software. And I must admit, I still use these clunkers to this very day, with no ill effects.

The QIC cartridge uses two internal reels of tape, a complex arrangement of capstans, and an externally driven friction wheel to move the tape. A "classic" QIC tape drive would use a fixed head with 9 tracks and employ simple linear recording. Capacity typically topped out at 150 MB.

Needless to say the QIC format became the subject of many a proprietary or would be standardisation and the number of different QIC variants has been listed as being about 120 types, a large proportion of which are mutually incompatible.

Limited to hundreds of Megabytes with unspectacular write speeds, it did not take very long for a new technology to knock the QIC cartridge off its perch. This technology was the 8 mm helical video tape cartridge which became the basis of the now ubiquitous Exabyte tape.

Helical tape technology evolved from VCRs, and employs a rotating head which is slanted relative to the tape path. As result, the data is not recorded linearly along the tape, but rather is recorded in diagonal stripes which run at an angle relative to the tape. While the linear speed of the tape is very modest, the head rotates very quickly, as a result of which the speed of the read and write elements relative to the tape is rather high. The read and write elements are paired on opposite sides of the head, and the data is written in a single serial stream which is staggered in consecutive diagonal stripes along the tape. This is a basic legacy of of its video technology origins, where video data is essentially a high speed serial analogue stream.

The 8 mm helical tape employed a block coding scheme, typically using a Reed-Solomon code and packaging the data into a matrix including a hefty proportion of ECC bits for forward error control. A late model Exabyte uses for instance a data matrix with 24 rows, sixty columns and 400 bytes of ECC per 1440 byte data block. Therefore about a third of the tape is used for error control.

The 8 mm Exabyte became the defacto industry standard for backups, providing initially 2 GB per cartridge which incrementally grew out to 5 GB with data compression.

The 8 mm cartridge tape was soon supplemented in the marketplace by the 4 mm Digital Audio Tape (DAT), often regarded to be a failure in its primary hifi marketplace. The DAT tape is like the 8 mm tape, a helical design but from the outset mostly used data compression hardware in the drive to achieve Gigabyte class densities with a much smaller cartridge. The data format for 4 mm DATs will typically be one of the DDS (Digital Data Storage) formats, most of which are nominally backward compatible in read operations.

While the DDS/DAT has become the defacto company standard of HP, it could not be considered to be the industry standard in helical tapes, since the 8 mm Exabyte type is very widely used and continues to be used.

More recently, we have the Digital Linear Tape (DLT) beginning to strongly proliferate in the marketplace, adding a third "defacto industry standard" into the fray.

The DLT is a late generation high density linear cartridge tape, using a very different cartridge format to the older QIC tapes. DLT drives employ a complex multilayered ECC and CRC strategy which is intended to provide a highly reliable long term backup medium. Since the tape motion is linear the head is fixed, and the mechanical stresses on the tape typical of helical drives are absent.

A typical contemporary DLT drive will record data in a superblock called a "data entity", comprising twenty 4 kbyte blocks, of which 16 contain data and four contain ECC. Each data block in turn contains fields with 16 bit and 64 bit Cyclic Redundancy Check polynomials, as well as an additional XOR based Error Detection Code (EDC).

It is illustrative to step through the recording process employed in a DLT drive to see how this is applied.

The placement of the Reed-Solomon ECC coding information in separate blocks widely separated from the data blocks is intentional, to minimise the odds that localised damage to the tape would knock out both the data and the ECC.

As is clearly evident, a contemporary tape drive is a vastly more sophisticated beast in comparison with the reel-to-reel tape technology of decades past.

Most manufacturers cite media lifetimes which are as great as thirty years, although in practice this will depend much on the storage conditions of the tapes and also on the number of write and read cycles, which stress the tape mechanically and cause wear to the magnetic medium.

A question which I am frequently asked is "how well can a modern tape survive an electromagnetic weapon attack". The answer is under most conditions, quite well. Some years ago I inherited a box of pre-loved 8 mm backup tapes, which the client asked that I erase before I walk out of the door with them. Easier said than done. The magnetic coercivity performance of the material was so good, that a degaussing wand placed on the tape for 2 minutes did not introduce a sufficient number of errors to prevent me from reading it back on the drive error free. Alas it was a busy evening feeding tapes into the drive and dd'ing /dev/null onto them !

Contemporary Tape Standards

The question of tape standards is a vexed one, in that a vast number of proprietary variants of the standard tape formats exist. In many instances these formats are backward compatible, in that a drive will read tapes cut by an earlier drive model, but can usually only write in its native format.

Interestingly, many older tape cartridge formats are still employed even if the data recording formats have been changed dramatically. The classical example are the QIC cartridges, which have spawned some interesting variants, many of which employ servo driven heads. Many variants employ a 0.318" wide tape rather than the nominal 0.25" medium and serpentine recording using a servo driven head.

The term "serpentine" refers to the fact that the data stream starts at one end of the record at upper side of the tape, runs to the end of the record, then back to the beginning of the record, and so on, thereby following a zig-zag or serpentine layout. The tape is run back and forth over the head, which is stepped along the tracks with each pass.

Thus we have a QIC-80 with 28 to 36 tracks and up to 500 MB capacity, a QIC-3010 with 40/50 tracks and up to 420 MB capacity, a QIC-3020 with double the capacity of the 3010, a 3080 with 60/72 tracks and up to 2 GB capacity and finally a QIC-3095 with up to 4 GB capacity.

These are now followed by the Travan 1, 2 and 3 standards, which provide 1, 2 and 3 GB respectively, and the new Travan NS providing 8 GB capacity with compression. Data transfer rates for some of these may reach up to a Megabyte/sec. The planned NS-20 and NS-30 variants are intended to store 10 and 15 GB respectively, or up to double these capacities with compression.

The DAT 4 mm tape is also generously endowed with alternatives to muddle the hapless user. The basic DDS standard provides 2 GB with a transfer rate of about 550 kbyte/s, the DDS-1 provides 2/4 GB at 0.55/1.1 MBytes/s, the DDS-2 4/8 GB and the most recent DDS-3 12/24 GB at 1.1/2.2 MBytes/s.

The current state of the art in 8 mm tapes is typified by the Exabyte Mammoth and the AIT standards. The Mammoth will fit 20 GB on an 8 mm cartridge at 3 MBytes/s sustained transfer rate, and up to double that with compression turned on. The drive is read compatible with older 8200, 8500 and 8500C standard tapes.

DLT drives of several years vintage would typically provide 20 GB capacity uncompressed, more recent models will exceed this.

Therefore should we choose any of the three established high performance tape formats, ie 8 mm, 4 mm or DLT, a 10-20 GB capacity is typically not a problem.

The vendor community continues to be active in expanding the choices a user has in tapes.

Of interest is the Linear Tape Open (LTO) standard produced jointly by IBM, HP and Seagate. The LTO standard is curious in that it defines two very different cartridge formats, intended for different applications.

The LTO Ultrium format uses a single reel cartridge, designed for maximum storage capacity.

The LTO Accelis format uses a two reel cartridge, designed for maximum access speed.

The data format is based on a linear serpentine model, using 8 and later 16 channels per head. Data is written using an RLL code. A servo mechanism is used on the head to maximise head positioning accuracy, and additional servo tracks are written adjacent to the data tracks. Extensive ECC is employed for robustness.

The Ultrium format is intended to store 100 GB on a 0.5" tape cartridge, and more in later variants. The tape has up to 384 tracks, split into "data bands". The head will be positioned by the servo onto the databand, the data written or read until the end of the block, upon which the servo positions the head to the next band and the tape is reversed. A head typically uses 4 or 8 read/write elements to read or write that number of tracks at a time. The cartridge has a coaxial layout, with a takeup reel coupled to the tape reel.

The Accelis format uses a narrower 8 mm tape, and two separate reels to achieve a higher linear speed than the Ultrium. Like the Ultrium it uses a servo driven head, with a total of 256 tracks split into two data bands. Basic capacity is 25 GB.

The long term outlook in tape drives is a continuation of the existing trend to produce denser media, more complex drives and retain some measure of read compatibility against older media in like cartridge formats. Other than that, the existing babel in standards is likely to persist, and tape drives will remain a proprietary domain for the forseeable future.

While we can expect the DVD to displace tapes in low end and single machine environments, advanced high performance tapes will continue to be unchallenged in centralised backup applications.