IBM Sets Areal Density Record for Magnetic Tape

digitalPhant0m writes to tell us that IBM researchers have set a new world record for areal data density on linear magnetic tape, weighing in at around 29.5 billion bits per square inch. This achievement is roughly 39 times the density of current industry standard magnetic tape. "To achieve this feat, IBM Research has developed several new critical technologies, and for the past three years worked closely with FUJIFILM to optimize its next-generation dual-coat magnetic tape based on barium ferrite (BaFe) particles. [...] These new technologies are estimated to enable cartridge capacities that could hold up to 35 trillion bytes (terabytes) of uncompressed data. This is about 44 times the capacity of today's IBM LTO Generation 4 cartridge. A capacity of 35 terabytes of data is sufficient to store the text of 35 million books, which would require 248 miles (399 km) of bookshelves."

Comments

In other news...

[Anonymous Coward]

... IBM researchers have set a record for compressing the most records of cattle onto clay tablets using their proprietary new cuneiform.

Re:In other news...

[afidel]

Are you saying tape is outdated? Because for organizations that have large storage requirements you can't get any cheaper than tape, and it has superior archival and transportation properties than HDD's as well.

Re:In other news...

[mlts]

The biggest issue with tape is keeping the density and cartridge capacity up. Even though tapes and magnetic hard drives share a lot in common, storing data on rust is where the similarities end. Tape media contacts the head and is read while the head is contacting, as opposed to HDDs which float very close to the surface of a disk. So, the material the magnetic domains are on has to be sturdy enough for physical contact. Tape does have an advantage that it has more space to store data than a disk platter. However, a lot of that space is taken up by error correction, since there is no way to relocate bad sectors on the fly with tape, and if a block goes bad, it goes bad, no way to recover without ECC, or a way to duplicate the lost data.

What tape has over hard disks is simplicity. A DLT or LTO-5 tape has one reel for a moving part. Compare that to a hard disk which has the platters, the heads, the wires, and the motors. Drop a tape, and it almost certainly is recoverable. Drop a hard disk, and a person never knows if the hard drive is completely dead, or will die very soon due to the impact. This is also important when it comes to archiving. Tapes, you can put in their cases, drop them axis vertical in an Iron Maiden tub, and your data is secure. With hard disks, you have to put them in padded boxes to help dampen vibrations which can kill the drive.

Tape drive makers are also responding to the clarion call of encryption. HP's LTO-4 line supports SPIN/SPOUT encryption capabilities. You can set it to use the same passphrase on multiple tapes, or use backup software which sets a different key on each tape and manages which key goes to which tape for better security. Software like Retrospect, Backup Exec, or bru also offer AES encryption with libraries certified by the US government. So, a tape backup is decently secure.

Tapes can be set to be read-only. This is important because it means that a tape read on a compromised machine won't be able to be tampered with. Some tape systems (DLT) offer WORM functionality to allow for secure archiving of data with the data cryptographically signed by the tape drive. This is important when one has to deal with HIPAA and archiving of data for 7 years, or the FAA and archiving airplane data for 50 years.

Tapes are fast. This is also one of their weaknesses. If you don't feed them the full amount of their pipeline, the tape drive has to stop and reverse. "Shoe-shining" is not good for tape life, nor the life on heads. So one needs to have tape drives preferably on a computer with the I/O paths to handle it [1].

Finally, once you buy the drive, tapes are the best bang for byte you can get. Even older tape formats like LTO-4 that give 800 gigs native for $40 is still fairly cheap for the capacity.

[1]: Ideally, the best use of tape is a network backup server with a good RAID array. You back the machines up to the array, then copy the data to tape. This way, network glitches do not slow down the data being slapped on the tapes.

You forgot to copy the stars (footnotes):

[Seth Kriticos]

*The demonstration was performed at product-level tape speeds (2 meters per second) and achieved error rates that are correctable using standard error-correction techniques to meet IBM's performance specification for its LTO Generation 4 products.

**Note that this calculation assumes a roughly 12% increase in tape length due to the reduced medium thickness.

***Note that this has been rounded up from 43.75 times

What about write speed?

[ircmaxell]

So what about speed? What good is the ability to store 35TB of data, if it takes you a week to write/read it?

Re:What about write speed?

[belrick]

If the areal density increases uniformly in both directions, then you can expect that the read/write speed goes up by the square root of the areal density increase. That is, for 43.75 you'd expect a speed increase of at least between 6 and 7 times. Note that sometimes the density increase is achieved in only one direction or the other, depending on what technology was used to achieve it, in which case all or none of the density increase results in speed increase.

You can achieve speed increases by using multiple heads. LTO and the 3590/3592 proprietary tape technology on which it is based use 8 or 16 tracks read/written simultaneously, with tracks interleaved. There might be 256 tracks with tracks 1, 17, 33, ..., 241 being accessed, then 2, 18, ..., 242. etc. Doubling the number of tracks (density increase of 2 widthwise) wouldn't increase read/write speed. Doubling the number of tracks while simultaneously doubling the number of heads would.

Note that with 8 or 16 heads spread across the tape width, error correction is achieved by writing a matrix of bits (across the tape as well as down the length) with ECC bits added.

Re:What about write speed?

[the_other_chewey]

... a quick google shows present tapes take between 25-50 passes to fill a tape at ~7 minutes per pass, god that's slow.

Slow? 35TB, 50 passes, 7 minutes per pass: 1.6GB/s (using decimal prefixes of course...)
I doubt it'll be that fast in practice, but slow it isn't.

In other news...

[seven of five]

Joe's Pizza Delivery and Data Courier Co loses the personal health and financial records of every human being on earth.

Re:AWESOME!

[timeOday]

Oh yeah. Assuming 400 MB per album (FLAC), a 35 TB tape will hold 87,500 of them. So, a million tracks, give or take. That should be fun to hunt through with nothing but FF and RW buttons!

Stupid Units

[Doc Ruby]

Why bother explaining how many miles of bookshelves would be needed to hold some amount of digital data? We don't explain how long a bookshelf would have to be to hold all the data in an HDTV screenful, and 35TB data tapes are probably going to hold more graphics than text. Besides, how big is the type in the books filling that shelf? And who but a librarian is going to relate to miles of bookshelves as a meaningful comparison, anyway?

Why don't they say "a 35TB tape is enough to hold 5 million full CDs, or 7,778 full DVDs? That's a comparison that people could actually relate to, that is actually factual, and isn't just some kind of primitive awe at how efficient we've become now that we store data on something not made of mashed trees.