Tape? Does anyone care about tape anymore?

Various friends and colleagues have been amused, occasionally stunned, but more often incredulous when I tell them that I am working on a project based on data tape. The commonly ventured opinion is that I should maybe think about working with a technology that someone has cared about in the past 30 years. ­čÖé

I understand where these people are coming from. In some ways my recent few years diving deep into data storage systems has felt like a backwards step from my time working on web-middleware, process-migration, and programming language work. But on the other hand, our need for data storage continues to grow at amazing rates. Ultimately continued expansion of data storage needs results in ever-increasing costs. So we, as a community, need to find ways to flatten out that cost curve.

As far as data tape itself is concerned, tape still has some very interesting properties for data storage. Tape is not the only answer for storage, but in many scenarios tape can be preferable over spinning disk or flash storage.

Some of the properties that make data tape storage interesting are outlined below the break. I’ll dive into more detail about these properties and usage scenarios in future posts where I explain how my work has made tape more accessible.

Longevity
Data tape media is very stable. Once the data lands on the media we can practically guarantee that it will be readable in 20 or 30 years. The media used for data tape is largely the same as was used in the 1950’s. Modern media has smaller particles, but the materials are relatively unchanged. We also know from experience that today in 2011 we can still successfully read tape media that was written in the 1960’s.
Reliability
Modern data tape like LTO is written using a head that is constructed with a write-head that is sandwiched between two read-heads. This head structure allows the tape drive to read back the data that has just been written to the tape no matter which direction the tape is moving. This read-after-write approach is used to validate that the bits were written to the media correctly. If a write-error occurs then the drive will mark that block as bad and re-write the data further down the tape. Various other safeguards around the media read/write operations such as ECC and aggressive error recovery are baked into LTO drive hardware
Reliability 2
Unlike hard drives, data tape is designed for the head to be in contact with the media. Also unlike hard drives, tape media is independent of the drive mechanism. These two ideas remove large classes of storage failure from being possible.
No head crashes since a head crash is ultimately a failure to maintain the necessary separation between head and media in hard disk storage. No problems with leaving tape media on a shelf and then spinning it up 5 or 10 years later. Hard drives tend to get cranky after sitting still for too long.
Robust
Tape media can safely be shipped safely. Ship a hard drive across the country a couple of times and you will end up with a dead hard drive. Surprisingly, shipping hard-drives (or in some cases shrink-wrapped pallets of hard-drives) around the world is common practice for video data. People who do this report extremely high drive failure rates.
Inexpensive
LTO5 media is roughly cost comparable to hard drive storage media. Based on experience with previous LTO media generations this cost will trend down quickly. At the hundreds of TB or higher space, tape and tape automation/robotics quickly become cheaper than spinning disk plus the disk interconnect.
Fast!
Individual hard drives top out with roughly a 100MB/s streaming data rate for writes or reads. LTO5 tape technology sustains a 140MB/s streaming data speed for read and write on incompressible data. Compressible data can be read/written at higher rates depending on the level of compression that can be achieved. (LTO data throughput is gated by the data rate at the read/write head to 140MB/s. If the data can be 2:1 compressed and a 280MB/s feed rate can be sustained to the drive then the hardware compression engine will compress the data and maintain a 140MB/s data rate at the read/write head.)
Green
Individual pieces of tape media do not require power for the media to be available online. Tape drives have very low power requirements when idle and are on the order of hard drive power consumption when operating.
In comparison, hard drive and flash storage require significant power and cooling (power also) if the media is online. At even modest storage scale, the power and cooling costs for hard drive and flash storage are a large part of the Total Cost of Ownership.
Seek Time
Data tape is a sequential medium, so seek times can be a concern. With LTO5, a full traversal of the tape involves traversing the length of the physical tape 80 times first forward, then in reverse, then forward, etc. But each “wrap” of the tape can be independently selected as a lateral move. Traversing the length of the physical tape takes around 90 seconds. So, a seek to any given block on the tape is, on average, around 45 seconds plus a couple of seconds for a lateral wrap change.A roughly 45 second average is obviously slower than a hard drive. But for large files, the long seek time is amortized by the 40% faster streaming data rate that LTO5 tape drives can sustain as compared to hard drives. 

This entry was posted in data storage, LTFS. Bookmark the permalink.

3 Responses to Tape? Does anyone care about tape anymore?

  1. Pingback: IBM and FOX win an Emmy for LTFS - IT Storage 411: Inside Systems Storage by Tony Pearson - IBM Storage Community

  2. Pingback: IBM and FOX win an Emmy for LTFS – Tony @ TechU

  3. Pingback: IBM and FOX win an Emmy for LTFS – Tony @ TechU

Leave a Reply

Your email address will not be published. Required fields are marked *