Recent Sales Hint That Tape For Storage Is Far From Dead

hightechchick writes "Staples' business-to-business sales of backup tape for storage are experiencing a bit of a revival. What's next, a return to dumb terminals and mainframes (a la cloud computing)?"

Comments

Not news.

[SatanicPuppy]

What else is there? It's not like you can back up to a SAN, and then stick the SAN in a courier bag and send it to remote storage. Optical? Too small. The magical "cloud" doesn't stack up well for security compared to a physical safe. Flash is promising, but still not there in terms of reliablity.

When they come up with a compact, reliable, portable storage medium I'll be the first one to toss tapes out the window. The idea of running backups to some credit-card sized SD cards is appealing.

Re:Not news.

[h4rr4r]

You could not be more correct. It also needs to be as fast as LTO-5 or LTO-6 when that ships. That means 140MB/s or 270MB/s, and at least close to it for long periods of time. Those cheap SATA discs the kids keep suggesting don't come anywhere near that.

Re:Not news.

[X0563511]

I (and many others) don't trust hard drives _at_all_ - let alone when you move them. This is learned behavior...

Re:Not news.

[Pinky's Brain]

Modern cheap SATA drives have average linear read and write speeds of around 120 MB/s.

Re:Not news.

[h4rr4r]

Please name a disk that can keep it up for the whole disk. This will be 1TB+ of random data written in 1 shot.

I have not seen any yet, but would love to find one.

Re:Not news.

[SatanicPuppy]

Yep yep. Too many moving parts. You can drop a tape, and 999 times out of 1000 it'll be fine. Hard drives? Hell, it could die of vibration damage in transit!

Tapes are small, disposable, cheap, reliable. Hard drives are maybe 2 of those.

Re:Not news.

[afidel]

Dude, my biggest problem is keeping the damn LTO4 drives fed at MINIMUM write speeds for file server type small file workloads. 72x15k spindles isn't enough with only one volume being backed up, metadata retrieval makes it too slow, I need to have multiple volumes backing up simultaneously to keep the things from shoeshining.

Re:Not news.

[Pinky's Brain]

If it's really important to you ... make a lot of partitions, RAID0 the lot of them ... hey presto, a volume which will maintain average linear read and write speed across the entire volume.

Re:Not news.

[sexconker]

You're an idiot, Starscream.

Tape is not legacy - it's the industry standard.
Believe it or not, being old is not the same as being obsolete.
In fact, in this industry, being old is a testament to how reliable something is.

Compression? Deduplication? Seeding remote sites? What fantasy world do you live in?

Tape is a storage medium.
You can compress anything and store it on the tape.

Deduplication is not a backup mechanism.

Backups need to be made before going live and routinley afterward. Full backups.

Tape is easy to restore from. You need full/incremental backups with tape exactly as you need them with a remote location. If it's attached to a machine it's a copy, NOT a backup. A backup must be remote, unpowered, and protected from Earth, Fire, Wind, Water, and Heart (thieves) etc.

It's not difficult to know if data is safe. Just try to restore it. If you're not testing your restore process, you're an idiot, regardless of what method you're using. Tape is the most reliable storage format we have today.

You can reuse tapes all the time. Such inefficiencies only matter if you're backing up data that's a fraction of a single tape. If this is the case, just buy more tapes. They're very cheap. If this is not the case, then you'll never run into the problem because each tape you write to will be part of a set of tapes corresponding to an individual backup job, and all but one of that set will be completely utilized.

It's not a needless nightmare. It's a necessary nightmare. And it's not a nightmare. There's this thing called a label maker. Alternatively, labels and a Sharpie. Alternatively still, tape, paper, and a pen.

No sir, it is you that deserves the beating.

Re:Not news.

[Just Some Guy]

Amanda works by backing up filesystems to dump files on the backup server, then writing those dump files all in one go. It might take an ancient system an hour to spool its dump to the backup server, but the tape doesn't have to worry about that.

Re:Not news.

[Culture20]

I fail to see how making partitions on one real device will do anything other than lower that average speed.

You are correct, it will lower the speed, but I believe GP is correct that it will bring most data points closer to that slower average.

If you meant multiple physical volumes in RAID0 then I ask the following:
And how do I get this volume off to the storage location?
Will the OEM say it is safe for transport?
Is it light enough for our female sysadmin to carry it?

1) Quantum Teleportation? Maybe a truck and packing foam if your teleporter is down.
2) Who cares what the OEM says? Are you planning to sue a tape manufacturer when a tape goes bad? Good luck proving it was the transport that did it.
3) Unless she's an invalid. Anyone who can lift a HDD can transport a disk array.

I say if JBOD backup with multiple copies is good enough to transport Antarctic science data, it's good enough for transporting backups. As long as the backups are tested at the storage site on arrival, then there's not a problem with disks.

Re:Not news.

[afidel]

Yep, but that's what it takes to keep it running above minimum speed, writing a single large filesystem volume takes longer than writing 4 volumes almost as large because the drive shoeshines with a single job. Minimum rate for the drive is 40MB/s, the best I have done with tuning on a 72 drive vraid6 volume is 35MB/s and average is closer to 25MB/s sustained but 4 jobs from the same array on different volumes will give me 100-120MB/s. All volumes are spanned across all disks so it's not a matter of more spindles being available, it's the latency in all the metadata lookups.

Re:Not news.

[turing_m]

Please name a disk that can keep it up for the whole disk.

http://hothardware.com/Articles/Definitive-2TB-Hard-Drive-Roundup/?page=7

You appear to be right. The best write average is about 100MB/s. It's the reads that are near 120.

Re:Not news.

[fuzzyfuzzyfungus]

I think you answered your own question there. Tapes are relatively cheap. Tape Drives are Not. Therefore, unless you are a large enough outfit to be amortizing the cost of the tape drive across a large number of tapes, tapes are effectively expensive. If you are, though, tapes are effectively cheap.

Re:Not news.

[hawguy]

Tapes are very cheap, LTO4 tapes hold 800GB (native, not compressed) and cost around $30.

It's the drives that are expensive for home user -- no home user wants to pay $2000 for a tape drive.

Re:Not news.

[cgenman]

The idea of backing up to a hard drive is just frightening. Hard Disk Drives are what you need to keep data safe FROM. I have things in my refrigerator that live longer than some of these damned disks.

Of course, I don't trust tapes either. But I don't trust tapes a lot less than I don't trust Hard Drives.

Re:Not news.

[tibit]

100MB/s is plenty. Consider the cost of the whole deal: interface card, drive, cleaning tape, aggravation of having to switch tapes/run cleaning job, and so on. Three external USB 2.0 drives beat that speed, and cost a whopping $420 at Walmart, of all places. You can go and buy them in the middle of the night if need be. Just plug them into the motherboard USB connectors, no recent enough system will have any problems keeping them all going full tilt. The drives will likely last 3 years of constant use. Now let's look at the numbers:

Assumptions:
- the tape has a lifetime of 300 passes, capacity of 1.5TB and $100 cost
- the drive lasts 3 years and costs $2000 (given my experience with LTO, that's optimistic)
- the SAS interface card and cable costs $300

- a USB external hard drive has capacity of 1TB and costs $130

Assuming that nothing gets any cheaper over time, you have to amortize the drive and controller cost over the $30/TB difference between tape and HD. You need $7500 worth of tapes to break even.

Now let's look at the lifetimes. If you're easy on tapes and generally lucky, a 10 tape set may outlast the drive. At that time, you may as well toss the tapes since you will want to buy a newer generation drive, and probably won't risk contaminating it with old crud from those tapes even if the drive may access them just fine. With my luck, half of the tapes had errors after one year, and that was on a lower density tape (VS160).

For LTO, I'd think that you'd want the backup will be written, then verified, and then any data that had errors will be written again. So the number of passes available from the tape's life shrinks by half to 150.

150 passes lets you use one tape to store and verify a total of 225 TB of data over its lifetime. How does that stack up to the hard drive? A USB 2 hard drive will transfer 3.9 TB per day, so if you operate it continuously it will outlive the tape after 2 months. Now of course noone uses backup tapes continuously, but that just gives you the idea of scales involved.

Now knowing that tape prices don't really drop much with time, in a year or two the hard drive cost per terabyte will be lower than the cost of a tape, and there's absolutely no reason to buy tapes any longer. With USB or eSata HDs, the interface costs are essentially nil, and most any current server comes with enough connectors to plug in several external drives at once, and the chipset is fast enough to keep them saturated if your source storage allows. You won't really be throwing multiple $2k tape drives at a bandwidth problem, but with USB or eSATA HD, the bandwidth comes essentially for free: as long as you have enough drives, each drive adds to the bandwidth. With tapes, a tape only adds to the storage capacity, but bandwidth costs $2k per 100MB/s!

So no, tapes make absolutely no financial sense unless your storage needs are immense -- as in a financial institution, media production, engineering/data mining, IT service provider... And even then, the hard drive technology is quickly catching up so that the slight win for tape drives is only a win for the next year or two. After that -- no excuse for sticking with tape, no matter what your size.

Re:Not news.

[Hognoxious]

Is it light enough for our female sysadmin to carry it?

Does it weigh less than 20 pairs of shoes (including retail packaging) and ten boxes of chocolates?

Re:Not news.

[KDR_11k]

Yeah but accessing your drive contents only through iTunes would be pretty clunky.

Re:Not news.

[WuphonsReach]

Tapes are small, disposable, cheap, reliable. Hard drives are maybe 2 of those.

The media may be cheap, but the drives are expensive and sometimes proprietary. So you'd best be a big enough outfit to buy at least multiple drives. Not to mention that you need to replace tapes regularly. At $2000/drive and needing at least three, plus needing 60 tapes per year at $30ea... you could buy around 30-40 1TB hard drives, with carry cases or trays. And you need to lay out that $7500 right at the start, plus the $1800/year. That's a lot of money for a small business with under 20 employees.

(And most tape drives are more like $3k to $4k each.)

The big problem with tape for smaller shops is simply up-front cost. For $150, they can buy a single 1TB drive and use that to write backups to. Each week, they buy a new drive until they are rotating 5 or 6 of them. Or if they really need to get data offsite daily, they'll do a delta-backup over the WAN links. Or spend enough to have 10 hard drives in rotation.

(We use a mix of backup over WAN links nightly/weekly combined with taking hard drives offsite weekly.)

Re:Not news.

[BikeHelmet]

Dude, my biggest problem is keeping the damn LTO4 drives fed at MINIMUM write speeds for file server type small file workloads. 72x15k spindles isn't enough with only one volume being backed up, metadata retrieval makes it too slow, I need to have multiple volumes backing up simultaneously to keep the things from shoeshining.

I've never done tape backups, but isn't that an issue with your OS? Maybe it's trying to compress data/files to save space?

With a linux LiveCD, you could probably dump a partition onto tape as quickly as you could read it from the raw device. It doesn't need to read the files individually, or understand the filesystem at all - it's all just data being read sequentially off HDD and being stored sequentially on the tape.

Re:Not news.

[totally bogus dude]

That's not much use if you want to be able to restore the individual files from the backup, which is nearly always desirable.

Disaster-recovery-only backups are okay, but if you're spending the money to archive your data you normally want a bit more flexibility.

Additionally there's the obvious problem of taking the server offline while you do the backup...

Re:Not news.

[totally bogus dude]

Points in favour of tapes:

- most backup software is designed to deal with tape libraries, not so much with shuffling B2D media around

- most archive companies are built around storing tapes; though I suspect there are ones which could deal with hard disks in external caddies

- tapes deal with stress from being transported continuously better than mechanical drives (also wear and tear of plugging and unplugging the interfaces all the time)

- I think unused tapes age better than unused hard disks, but I've nothing to back that up

Bandwidth to the tape drive itself rarely seems to be an issue for actual backups, since network and file I/O latency seem to be more significant issues. We never get anywhere near the maximum speed out of our LTO-4 drive, even when we're just duplicating data from the local array to the tape.

Re:Not news.

[jimicus]

I actually re-read this post several times because I wasn't too keen on the tone. I'm not having a go at you, but I was running the numbers myself on a spreadsheet only about a month ago and was expecting similar results to what you suggest.

I was totally wrong. Right now, for any non-trivial quantity of data which is expanding at any non-trivial rate, LTO4 is the most cost-effective solution. (Actually, the most cost-effective solution is probably an LTO-5 drive but use LTO-4 tapes until the LTO-5 tapes come down a bit in price).

100MB/s is plenty. Consider the cost of the whole deal: interface card, drive, cleaning tape, aggravation of having to switch tapes/run cleaning job, and so on. Three external USB 2.0 drives beat that speed, and cost a whopping $420 at Walmart, of all places.

Please explain to me how you maintain 100MB/s write speed on a bus which can - on a good day - manage just under half of that. And usually manages more like about a third.

Assumptions:
- the tape has a lifetime of 300 passes, capacity of 1.5TB and $100 cost

Or you could buy LTO4 tapes which will write just fine in an LTO5 drive and right now are quite a bit cheaper per gigabyte.

- the drive lasts 3 years and costs $2000 (given my experience with LTO, that's optimistic)

Are you not getting a 3 year warranty on your drive?

Assuming that nothing gets any cheaper over time,

Really? I'm buying LTO3 tapes for about a third what I was paying a couple of years ago.

With my luck, half of the tapes had errors after one year, and that was on a lower density tape (VS160).

Seriously, if half your tapes have errors after just one year, there is something seriously wrong. I don't know if it's power, environmental factors, cheap tapes or what but there is no way you should see a failure rate that high.

For LTO, I'd think that you'd want the backup will be written, then verified, and then any data that had errors will be written again. So the number of passes available from the tape's life shrinks by half to 150.

You are aware that the LTO specifications include automatic verification as part of the writing process? You generally can't turn this off.

150 passes lets you use one tape to store and verify a total of 225 TB of data over its lifetime.

Splitting hairs, but LTO writes a full tape in several passes. A tape will last several thousand passes, but probably only about ~150-200 complete fills.

Pictures explain this far more clearly than text at http://www.lto.org/technology/primer2.html

How does that stack up to the hard drive? A USB 2 hard drive will transfer 3.9 TB per day,

Unless your users will put up with the performance hit that comes from taking backups during the working day, it doesn't really matter how much you can transfer per day. What matters is how much you can transfer during your backup window.

Now knowing that tape prices don't really drop much with time,

Which is wrong.

With USB or eSata HDs, the interface costs are essentially nil, and most any current server comes with enough connectors to plug in several external drives at once, and the chipset is fast enough to keep them saturated if your source storage allows. You won't really be throwing multiple $2k tape drives at a bandwidth problem, but with USB or eSATA HD, the bandwidth comes essentially for free: as long as you have enough drives, each drive adds to the bandwidth.

Not true, you'll be limited by the bus speed very quickly indeed. Just because your system has 8 USB ports does not mean you can expect to see 8x480Mbps when you've got 8 hard disks plugged in.

With tapes,

Re:Not news.

[BobMcD]

That's not much use if you want to be able to restore the individual files from the backup, which is nearly always desirable.

Disaster-recovery-only backups are okay, but if you're spending the money to archive your data you normally want a bit more flexibility.

Additionally there's the obvious problem of taking the server offline while you do the backup...

If you're pulling individual files off of tape, you're probably doing it wrong.

Backup across the network, to disk, first. You can build or buy a wide variety of arrays to do this for less than your tape drive costs, on average. Go large and rotate the storage mount points. We keep five days 'on line', and overwrite by schedule.

Write THAT data to tape, to be sent offsite.

On the upside, you can get any file from the last five days in less than an hour, without leaving your desk. More like fifteen minutes, really. Disks are for retrieval, tapes are for archive and disasters. Very clean, very simple, auditors love it.

On the downside you've doubled your costs, have additional overhead, and are probably adding lag to your tapes by extending the time-to-tape by a full day.

Still, though, if you can swing it, do.

Offsite backups...

[nweaver]

Although disk is compellingly cheap, if you want reliable, multiple, and offside-stored backups, tape really is the answer.

Re:Offsite backups...

[h4rr4r]

Single disks are slow as hell too. Try seeing what a single cheap SATA disc can sustain for writes, they suck.

What's Next?

[Monkeedude1212]

I'll tell you whats next...

A device that lets you type up a word document and it prints it out real time. Essentially it prints your keystrokes as soon as you press them. I also foresee it having an extremely long battery life.

Mainframe and tape

[tooyoung]

I'm guessing this story was posted by someone with absolutely no experience with enterprise-level businesses.

Re:Mainframe and tape

[icebike]

I'm guessing the story was posted by someone who does not know how to post a proper link.

The actual story is here:

http://www.channelinsider.com/c/a/Storage/Tape-for-Storage-Staples-Says-Tapes-Demise-Greatly-Exaggerated-339951/

Re:Mainframe and tape

[confused one]

I'm guessing this person(s) also has no experience setting up a disaster recovery plan with offsite storage for a small to medium sized business.

Real link

[phantomfive]

Here is the real link that is missing from the summary.

I always wonder about tape backup.....it seems everyone I know who uses it has had it fail. Hard drives fail too, it's true, but the anecdotal evidence I have says if you are using tape backup, you better have multiple backups.

Re:Real link

[h4rr4r]

How does that matter since disks are not even in the running? They are slow, not safe for storage, not safe to transport.

Real world advice from an enterprise sysadmin:
If you are doing backups, always have multiple backups since a single one will always fail when you actually need it.

Re:Real link

[TooMuchToDo]

The problem with tape vs live backups is that you can immediately test the integrity of your data against checksums or other redundant chunks of the data in realtime. Not so with tape. Once you write to tape, unless you check it every so often, you have no idea if the data is still good or not.

Trust but verify.

Re:Real link

[h4rr4r]

Live backups fail the minute you have a real big failure. Unless you have multiple sites and a huge pipe between them. Then you still should have tapes off site so one employee can't go destroying all your nice backups when he goes nuts.

Re:Real link

[h4rr4r]

You check the tape at the end of writing it. Everyone does this. You also test your backups, if you are not testing them you do not have backups.

Re:Real link

[icebike]

Tape sucks.

I've used it for 25 years, with a variety of vendors, capacities and dozens of drives, and every single unit I've ever had failed, not only the tapes themselves, but also the drives. People can't remember to cycle tapes, tapes die and people don't notice, and you can't buy the tapes 3 years down the line.

Disk is much simpler, and more robust.

We finally realized that we were backing up a very reliable media with a very un-reliable one.

Finally we switched to compressed backups stacked on cheap redundant network attached disk drives in small external enclosures. They can sit anywhere, even INSIDE the fireproof vault.

The software for this is readily available from a number of sources and you can use your same "tower of Hanoi" media cycling schemes as you might for tape. Because backups are bundled into one large file they can be stored, cataloged, archived, rotated, and purged via automated means.

For the small business, NAS drives make way more sense.

Re:Real link

[TooMuchToDo]

Example: My last gig was with the DOE working on the US site for the CMS experiment for the Large Hadron Collider. We had around 5PB of spinning disk and 17PB of LTO4 tape storage for the detector data (you can't really backup 17PB offsite for a reasonable cost). We'd have bad tapes quite often, and it didn't matter if you did a verify at the end of the tape write before it was stored by the robotics.

Re:Real link

[TooMuchToDo]

We did this at my last gig. You'd still have bad tapes, didn't matter if you checked the tape at the end of the write.

Re:Real link

[PhunkySchtuff]

Oh, you go and store 17PB on LTO and you think you know it all do you?

I'll just duck off to my corner and resume setting up my 8-slot autoloader.

As you were...

Re:Real link

[icebike]

Get a clue.
This article is not about enterprise. Its about small business. Otherwise Staples would not be involved.

Re:Real link

[TooMuchToDo]

While the data isn't backed up all in one spot, it can either a) be reconstructed from other data, b) regathered from the 800+ other facilities we distribute chunks of the data to, or c) recollected. It's cheaper than the $8-12 million it would cost to backup all 17PB offsite (and that's taxpayer money).

Re:Real link

[TooMuchToDo]

More than 17,000, all stored in huge Storagetek libraries:

http://upload.wikimedia.org/wikipedia/commons/e/eb/Storagetek-tape_drive_hg.jpg

More info on CERN's infosystems for the collider, as they're the Tier-0 site (which means, in realtime, they take the raw detector data, strip it to the bare essentials, and than shove it out to Tier-1 sites at up to 40Gb/s (depending on the detector/experiment):

http://news.cnet.com/8300-11386_3-76-2.html?keyword=CERN

A revival?

[Sycraft-fu]

Did it ever go away? As far as I knew it was always how you did long term backup. We just bought a new tape unit here since we needed more backup storage, and the development of new LTO formats continues apace. Disks are what you use for online storage, and for online backups, you have redundant disks. That is for sure our first line of defense. We have a RAID-6 system with hot spares. Ok, but what about if something bigger happens? I'm not just talking about facilities destruction, what happens if something goes apeshit in the storage system and screws up all the data (or maybe a malicious admin does)? If our backup is just a realtime hookup to another online system, we are screwed.

Tapes though, the protect for a lot of things. We take regular backups, in rotation, so that even if the online system is messed up, there are backups to go to. Those backups can also easily be rotated to secure storage facilities. These are places that aren't easy to have an online system, even if you wanted. You are talking like a vault or something to keep it safe even in extreme situations.

They are also great if we want to keep data for a long time. Tapes have good shelf life. Better than a HDD. This is largely because they are simpler. They are just, well, tapes. Retension them once a year, they can last decades.

So I wasn't aware tapes had gone anywhere. We sure don't use them on individual machines, or use them as fast backups, but they are wonderful as an emergency backup. The protect against a number of issues that an online disk system can't. They can also easily give you the benefit of offsite backups for a much lesser cost. Costs a lot more to get a second high end storage system and house it in another building with fibre than to just walk some tapes over to a vault.

Let's not forget Escrow

[frooddude]

The business I work for goes through tapes like they're used to make coffee. Primary use: legal escrow of source code.

Sheltered geeks

[Vrallis]

This is why us geeks on the 'working end' of the spectrum hate dealing with the sheltered back-end IT geeks.

Yes, we use tape. It's portable, easy to swap, easy to use, cheap to replace when it wears out, etc.

I do have cases where backups are made to disk or over network--then those backups go to tape so they can be rotated offsite.

The one case where I'm stuck dealing with backups to a portable HD between Windows, VMware and the backup software in question the whole setup is so badly broken that the entire thing has to be rebooted in order to swap the USB hard drive to rotate it offsite.

The people responsible for making comments like 'tape is dead' need to be dragged (probably kicking and screaming) into the real world for a while and learn what all their toys are really used for. A server handed to us with a fresh OS is just a doorstop until we actually get applications on it and it is actually capable of *doing something*.

Okay, done ranting for the day...

IBM Long Term File System

[paranoidd]

IBM recently announced LTFS (Long Term File System), which allows one to operate LTO-5 tapes as if they were a normal file system.

That's a very exciting technology which allows for the standardization of tape formats -- its specs are freely available in the LTO Consortium website and the implementation has been released under the GNU LGPL (see the LTFS website for links).

Tapes are not dead, certainly!

LABEL them FFS

[bsercombe72]

One thing that is generally underdone in our industry is proper labeling of the backup media. You NEED: software that wrote the backup including version Date Tape number in the sequence for this backup (tape 1 of x) hopefully a brief description of what was backed up. In 8 years 10,000 tapes in archive boxes with nothing but barcodes is pretty useless. The catalogues no longer exist.

Paper or plastic?

[quenda]

For *real* archiving, you cannot beat paper tape. I don't trust it unless I can see the holes.

Disk to Disk Backups

[anexkahn]

Or organization uses Disk to Disk backups. Meaning we backup our SAN to a secondary SAN, then that secondary SAN replicates to an offsite SAN. So we end up with 3 copies of our data.

The problem with Disk to Disk vs Tape: One day someone who didn't understand LUNS mounted the same LUN onto multiple servers and the two servers managed to clobber a bunch of the backups on the secondary SAN. Its harder to do that with tape....but tape has it's own issues.

My recommendation? Disk to Disk to Tape. Use the backups you store on your disks to do your quick restores and use tapes as your off-site backup. It is probably the most cost effective solution, and since the tapes are not plugged into anything no one can touch your server and instantly wipe out ALL of your backups.

Re:I was about to mod you up. . .

[hey!]

Not deserving a beating is not the same as not needing one.

Unfortunately, people remember negative experiences more clearly than positive ones.