Tale of the Tape: Beware of Wind Quality
A little more than a year ago, a customer told me his company was going to write tapes remotely over an OC-3 connection. The tape drive being used was the StorageTek T9940B, which has a native transfer rate of 30 MB/sec and can support up to about a 68 MB/sec transfer rate with compression. The average compression this customer was seeing was about 1.5 to 1, so he was running the local tape drives at about 45 MB/sec. After factoring in packetization and contention, we figured that we would be very lucky to see 15 MB/sec on this OC-3 line, and would likely see about 6 MB/sec. I cautioned them that this was a bad idea.
Tape wind quality can be a problem on local Fibre Channel networks, but the problem is exacerbated when replicating directly to tape remotely.
In this installment in our series on data replication, we hope to shed some light on the little known issue of wind quality posed by high-end tape media (LTO-I, LTO-II, IBM 3590 and STK 9840/9940). Lower-end tape can have the same problem, but more often than not, since the tape is running at a slower speed, users can keep up with the tape drive performance.
In the case of tape wind quality, a picture really is worth a thousand words. Here is a picture of good wind quality:
|Photo 1: Good wind quality|
As you can see, the above tape is flat and there are no issues with the tension of the tape.
Here are two pictures that show poor and marginal wind quality, respectively:
|Photo 2: Poor tape wind quality|
|Photo 3: Marginal tape wind quality|
Over the long term, the data on the last two tapes has a much greater potential to have problems with data reliability. As you know from previous articles (see Preparing for a Disaster and Preparing for a Disaster, Part 2), on a per-device basis, high-end tape has about two orders of magnitude lower bit error rate than Fibre Channel disks and about three orders of magnitude over SATA devices, but all that changes with poor wind quality.
Almost all tape drives are designed to stream data to the tape. If the drive cannot be streamed, it constantly starts and stops. This start-stop action combined with the tape threading within the drive can cause the tape to wind on the reel improperly, creating "scattered" winds. This happens because when the data stream is interrupted, the drive must stop and back up to reposition the read/write heads before starting again. This results in slight tension transients that allow the tape to wander ever so slightly. This is true for every drive type and vendor, be it AIT, DLT or LTO. It is not an issue of the quality of the drive hardware or media (although high-quality media can reduce the occurrence), but a fundamental issue with all tape drives when two smooth surfaces come together at high speed.
Those familiar with magnetic tape have seen this problem since the days of reel-to-reel tape, and knowledgeable media experts have warned about the issue for years. Over time, the compressive forces within the tape pack will result in plastic distortion of the tape edges associated with these scattered winds, and increase error occurrence.