Can you determine the quality of a cartridge?
Unfortunately, it's difficult to determine the quality of a tape cartridge. Most users don't have the resources...
required for an ongoing, full-scale quality assurance program, even though the benefits of such a program are obvious. The only method available to the general user is to monitor the read/write errors of the cartridge. Read/write errors can be caused by the cartridge as well as by the tape drive. As a cartridge becomes used, an increase in the number of read/write errors could identify a deteriorating cartridge. LTO drives keep internal track of read/write errors; some drives gather fairly detailed statistics. But usually, only the drive manufacturers have ready access to these detailed statistics. Some of the read/write error statistics assembled by the drives can be accessed through the interface (using the log sense command). User software can track cartridge quality, and can identify a potentially deteriorating cartridge, hopefully long before disaster strikes. In addition to using the log sense command, users may also obtain error statistics by accessing the cartridge memory chip inside the LTO cartridge. A special reader that can talk to cartridge memory chips is needed. LTO drives use such readers to update the cartridge memory every time the cartridge is loaded into the drive. But they aren't widely available as separate units. For more information on separate unit cartridge memory readers, visit http://www.mountainengineering.com/cm.html.
Choosing among the competing archival tape technologies is no easy task. For many storage managers, the technology of choice is LTO. However, not only are there three companies which manufacture LTO drives--Certance, Hewlett-Packard (HP) and IBM--there are also four major vendors of LTO Generation 2 media--Fuji, Imation (also sold under an IBM label), Maxell (also sold under an HP label) and TDK.
Once you decide which drive to purchase, the next big decision is: Which media should you use? LTO media quality, which can vary, is extremely important to successful backups and restores.
We randomly purchased two cartridges each from four LTO media manufacturers: Fuji, Imation, Maxell and TDK. All cartridges were manufactured in 2003 and are 200GB Generation 2 cartridges. Our tests proved that not all LTO tapes are of the same quality. All LTO cartridges usually work in any LTO tape drive, but the tapes--which are all manufactured to the same specification--vary considerably. This leads to the question: How do you evaluate LTO tape cartridges?
Just a few of the many parameters that determine the overall quality of a tape cartridge are: the reliability of the cartridge mechanism; the edge quality of the tape; the consistency of the magnetic layer on the tape; and the accuracy of the servo tracks. We will show why the quality of the tape edge is so important to successful backups and restores and also describe what our tests of leading manufacturers' LTO media revealed.
Tape edge bumps
Tape edges are produced during the manufacturers' slitting process. Slitting machines slice the original tape, which is produced on wide rolls, into the traditional half-inch tape. Each roll produces many half-inch tapes. However, slitting really ought to be called tearing; in reality, the tape actually tears in front of the slitter. The knife tips in the slitters never touch the tape itself.
It's the same phenomenon that occurs when cutting a long piece of paper in a straight line using scissors. Other than pushing the scissors forward, hardly any movement is required because the paper is tearing ahead of the blades. Not even the most meticulous manufacturing processes have been able to avoid producing rough edges. Viewed microscopically, these torn tape edges appear spectacularly uneven.
To make things worse, data tracks on an LTO cartridge are narrow: roughly 20 micrometers on a Generation 2 cartridge (about one-fifth of the thickness of a sheet of copier paper). As tape is moved over the read-write head in a horizontal direction, it inevitably moves up and down because of its rough edges. These vertical movements can grossly disrupt the crucial alignment between the data tracks and the read elements. It's a constant challenge for a tape drive to properly align the narrow data tracks with the read elements in the read-write head.
The LTO drive usually solves the track-following problem in two ways: First, the tape is pushed down so the lower edge of the tape rests against several reference guides along the tape path. Some tape drives are more successful than others in preventing vertical tape movement in this manner, but none are entirely successful.
The second method of track-following involves mounting the write-read head on an actuator that shifts the head vertically with the tape movement. A special read element in the head, called the servo read element, looks for prewritten tracks on the tape called servo tracks. Servo tracks are written by the cartridge manufacturer during the manufacturing process and are specifically for track-following purposes. As the tape moves vertically, this movement is detected by the servo read element. The actuator will then move the head vertically in an attempt to follow the tape.
Tape-edge distortions can be compared to hitting an unexpected pothole while drinking coffee in your car. The resulting spilled coffee is analogous to data loss due to tape edge distortions in a tape cartridge. But the tape is moving at a speed of roughly 14 miles per hour, and its microscopically uneven lower edge is sliding against the reference guides. If one of the jagged bumps on the tape edge encounters the reference guide, the tape will naturally bounce up and down fairly erratically. Meanwhile, the actuator is trying to follow the swiftly moving servo track throughout all this excitement--but with limited success--and the edge irregularity causes the tape drive to lose some data. Clearly, the best tape to buy is one that has the fewest bumps and bumps that don't slope too steeply.
Slope and amplitude of edge bumps
All eight tape cartridges we tested had bumpy edges, but they differed in both the height (amplitude) and the steepness (slope) of the bumps. The graph shows the standard deviation we measured for both parameters; in general, the shorter the bar, the better.
Testing the tapes
We measured the quality of LTO tape edges with a special tape transport developed specifically to test tape edges. (see "How we tested," on this page). Tape is moved from a supply reel to a take-up reel under the control of a servo system. The tape is guided by ten precision air bearings. These air bearings push down tape gently but firmly against a long reference edge. Only the peaks of the tape edge bumps touch the reference edge; the tape doesn't move in a lateral direction. (For more information about the test tape transport, visit http://www.mountainengineering.com/transport.html). A precision optical sensor is positioned in a gap on the reference edge. This sensor measures the amplitude and the slope of the edge distortions, or speed bumps. The diagram called "Up close, tape edges are rough" on this page shows the location of the optical probe in a gap of the reference edge. Only a narrow section of tape is shown. Under magnification, tape edges take on a mountainous appearance. The magnitude of tape edge distortions is typically between three to 10 micrometers.
All tape edges have distortions, but two variables are the most important: the size and the slope of the distortion. The larger the distortion, the greater your chances are of losing data, but the slope of the distortion is also important. A gently sloped bump will do less damage than a steep, abrupt one. Both the amplitude and the slope of the distortions are significant, so smaller, gently sloped speed bumps cause less damage and data loss.
To compare tapes from different manufacturers, we needed to have a single unit of measure for our speed bumps, one that would quantify their amplitudes and their slopes for an entire tape. We first calculated the mean and the standard deviation. The deviation statistic produced an excellent unit of measurement for an entire tape. From there, we could easily compare the standard deviations that we derived from the different tapes we tested. Secondly, we calculated the frequency content of the slope, which is the steepness of the bumps. Also, for ease of comparison we calculated the standard deviation.
The graph called "Slope and amplitude of edge bumps" on this page shows the standard deviations for the amplitudes, or heights, of the irregularities in our eight test tapes. A larger number indicates both a greater frequency as well as a greater size of the jagged edges. This graph also shows the standard deviation for the slope of the bumps. A small standard deviation indicates that they are gentle; a larger number indicates a steeper slope. Notice that the two graphs match fairly closely with respect to LTO media manufacturer. Tapes that had a higher frequency of edges--and more that were jagged--also tended to have steeper bumps.
As mentioned earlier--and with respect to this single aspect of tape quality--the highest quality tapes were those that didn't have too many bumps, and those that did exist were fairly small. Also, the slope of those bumps was not too steep. Do these results suggest that you should only purchase tapes from TDK because one of its tapes showed the lowest amplitudes or that you should purchase tapes only from Fuji because one of its tapes showed the best slopes? Not necessarily.
Remember, we only examined one indicator of tape quality--the tape edge--and our test sample was small. A different test, or perhaps the same test repeated with different tape samples, may produce vastly different results. But in the absence of any other data or any other means of testing edge distortions, we trust these results more than we would the sales spiel from a tape vendor.
In addition to the tests described, we also looked at many other tape edges and made an unexpected discovery: All the tape edges had the most damage at the beginning and the end of a tape. This was surprising because we had originally associated tape edge quality only with the slitting process, at least as far as new tapes were concerned. Because slitting is done on large rolls of tape--before the tape is cut into the cartridge size length (about 600 meters)--why would the opposite ends of the tape show more edge damage than the middle?
We concluded that this damage was caused by the equipment that spools the tape out onto the cartridge hub. This equipment attaches the end of the tape to the hub, and it also adds a leader to the beginning of tape. In all likelihood, it's this process that's causing further harm to the tape edge. Not all tapes showed the same amount of damage, but all suffered some damage. We think that there's room for improvement in this area.
Obviously, more conclusive testing would require a larger sample size, which would also produce higher quality test results. For instance, we only tested two samples of tape from each LTO tape vendor. The edge quality may fluctuate with changes in the manufacturing slitting process, so tests should be done on an ongoing basis.
Furthermore, we only ran tests on new tapes out of the box. Are the LTO drives really "tape-edge eaters," as has been rumored? To determine if there's any truth to that statement, we would need to test a new tape, run it in an LTO drive and then retest it after 1,000/5,000/10,000 passes to see if its edge further deteriorates. Finally, we tested only LTO tapes, but the same or similar tests could be done for other tape technologies as well.
Testing tape quality isn't easy, but it certainly can be done. Because a considerable proportion of a budget is regularly spent on purchasing LTO tapes, investing some resources to discover the quality of LTO tape would be money well spent. We believe that anyone who purchases tapes in quantities should seriously consider setting up a program that periodically tests not just the tape edge quality, but also tests other tape quality parameters such as the integrity of the servo track, the number of media defects and the mechanical components of the cartridge.