Shrinking Tapes And Increasing Bit Densities?

MHQ13 writes: "Over the years tape media has advanced greatly, increasing in bit density and decreasing in tape length and physical size while increasing in total storage capacity overall. Although I feel this is a good thing, I am not satisfied. Why are manufacturers not producing longer and larger tapes with current bit densities so that we can have tapes with monstrous total capacities? I would like to have tape media capable of storing a terrabyte or more. This would greatly simplify the nightly backups that I have to perform to backup our numerous UNIX systems. I think it would be a great benefit to be able to perform full backups of every partition of each machine either every night, or at least 2 or 3 times a week, rather than having to perform primarily incermental backups and squeezing in a full backup about once every two weeks."

Monstrous tapes

[He Who Suffers]

Seems like a good idea to me too, but I'm sure that the tape drive manufacturers would like to sell more stackers for their existing range of drives.

What about tape changers?

[Kris_J]

The last place I worked had an 8-tape DDS2 tape changer. 32GB native, 64GB compressed. If a similar system was available for DDS3 tapes you would have 96GB native, 192GB compressed. It's not the terrabyte you want, but honestly, a lot of companies don't need that sort of storage. We've only got about 32Gig on our network.

Big tape = big headaches

[toybuilder]

Bigger/longer tapes have lousier yields. Just like IC's and LCD's, as the surface area increases, the cost of testing and the probability of excess defects increase until it's no longer cost effective to manufacture and sell.

Next, your bigger tape will wear out faster than its smaller brothers because there's a lot more tension in the spools. Sure, the manufacturers find thinner substrate to fit more "linear meter/feet" of tape in the cassette... But this only exacerbates the tension problem even more.

Making wider tapes? Well, now you're talking about increased head size and higher cost (because the head is bigger, or requires more sophisticated control of head-travel geometry).

Besides, do you really want to store all your eggs in one (Terabyte-sized) basket?

AIT-2 Autoloaders

[mjdl]

Try tape autoloaders using Sony's rather expensive AIT-2 format: 50GB (uncompressed) raw capacity per tape. Small format, so a smaller capacity autoloader cartridge can be easily couriered off-site.

Some AIT-2 library links for more info:

• Sony's autoloaders [storagebysony.com] (4 tapes per cartridge, 1 drive).
• Tapelibrary.com's offerings [tapelibrary.com] (12 to 360 tape libraries, up to 12 drives).
• Of course, for true storage freaks in all formats, nothing can beat ADIC [adic.com] libraries, and while their higher capacity AIT libraries [adic.com] (up to 46,656 tapes, 256 drives) are not exactly portable, they do have more reasonably sized machines as well (11 tapes, up to 2 drives).

Anyway, the point is that solutions do exist for those people worrying about the expanding MP3 collections on their drives and all that money in their bank accounts...

Timing...

[JacobO]

It strikes me that the biggest problem is in how long it takes to back things up. Not the capacity. Going with multiple DLT drives and some kind of library/stacker is a nice solution. I personally wouldn't want data from 50 servers all going onto one tape. Not to mention how long it would take to do it. Getting the data to the drive fast enough is the problem.

Re:What about tape changers?

[KyleCordes]

Such a changer costs several thousand dollars, correct?

A home computer costing much less than that often has more than 32G of storage. A server with a small disk array can have a LOT more than 32G of data to back up.

Re:What about tape changers?

[spiro_killglance]

Our company just brought a 6 tape DDS-4 Drive to backup all 20 or so PCs. At £1500, and with 240GB native i consider that very good value for money. I

convience for robots.

[bluGill]

I work for STK, the king of high end tapes. Most of our customers report that the majority of their tapes are less then 30% full. Unix can use a tape drive only for backup, the big money in tapes is in the mainframes which uses tape for a lot more. To a mainframe you use cheap tape (and when your data center houses 300+ terrabytes tape is significantly cheaper then disk) to store data for user programs without hitting disk. Not in all cases of course, some things need high speed access to everything, but those jobs that you don't need to quickly generally use tape for their storage.

Becuase tape is generally unformated there is a much higher bit density, but there is no random seak on write. You can write the whole tape, or you can advance well byond the end of the last data section to make sure you don't over wright something. Because of the danger of starting before the end of the last section few people bother to write two different data sets to a single tape.

Don't overlook access time either. Tape is liniear, if you need something at the very end of the tape you need to read the whole tape to get to it, with small tapes that can already take several minutes. Do you really want to wait? Are you sure? Backups are nice and all, but if you don't restore from them what is the point? If a single tape that holds a terabyte would work for you, then so would a good RAID-5 (or mirroring) Disk system. If you want your backups to not only deal with disk crash, but also human error deletin a needed file, then tape is the solution, and smaller tapes make it eaiser and faster to get to the one accidently deleted file.

If you are having problems manageing your tapes, our salesmen will be happy to sell you a multi-million dollar robotic library. You should be investigating these systems as the better solution to your problem. I would guess that a smaller system would serve your needs just as well, from either us or our compitition.

In other words I think you are solving the wrong problem, and so you have come up with the wrong solution.

Re:Yes, RAID is not the answer

[Chanc_Gorkon]

Correct! Mainframes have had RAID LONG before PC's and servers. RAID is only good for HARDWARE failures. This is why a drive can fail and we stick a new one in using hot swapable drive and you can just insert another drive and it handles the problem. A lightning strike WILL take out not only your RAID array, but the server as well. Lightning can affect you even if you are on conditioned UPS power, but usually nothing as severe as just running on a surge protector. Usually it would just kick powerchute or whatever your using and have it down the machine, but it's still an inconvenience. Tape backups are still the cheapest way of off site storage. Things are impoving though. With end to end broad band connections, some installations are using a mirroring type system where the remote system will mirror your system over IP or a secure T1. This way all you'd have to do is get the new system built, then connect and restore all of the data, or if your really fancy you never went down anyway even if the main center took a hit (automatically cut over to the hot site).

Re:convience for robots.

[dmatos]

Um, not to be insulting or anything, but did you read the question? The tapes are being used for backups, not to store data necessary for running programs. Since they are solely for backups, I'm sure waiting a few minutes in the necessity of a restore would not kill anyone. Access time is not his problem, but tape capacity is.

When I did a co-op stint in IT at a manufacturing plant, we did nightly backups of our servers. It was basically one tape per server. Luckily, we only had about five, so that was five tapes per day. RAID-5 is not an answer, because the tapes were not only in case of disk crash/user error, but also fire, flood or other catastrophe. Our backup tapes were kept at an offsite secure storage location for this reason.

As for users, if they are clumsy enough to delete an important file, they deserve to wait to get it back. Using our system, it could take up to three hours to recover a file from backup tape. It just taught our users to be a little bit more careful.

IBM ADSM now TSM (Tivoli Storage Management)

[Chanc_Gorkon]

That's what we run. Big magnastar tape library (well, R2D2 sized anyway) with a RS/6000 server for running the TSM software on AIX. All of our servers backup to a RAID array, then at scheduled times, the RAID will be backup by tape. In an ideal situation which we don't have you'd have one generation on the RAID, and 2-3 on tape, at least one of which would be offsite. Right now we don't have enough tape to do it. It's spiffy once we get enough tape. It also has a client for almost every OS imagineable, even Linux and Windows included! As it stands now, we need to get another library to hold the tapes because we were already short and we just add 36 GB of expandable SAN from EMC! Those are going to be the main application volumes for the servers just installed. Each server only has one disk in a non raid stiuation (nothing of value is on that drive, just the boot code), which will probably be backed up to tsm as well. Big tapes ARE coming and they aren't all bad so far as access time. If you want REAL good access time and tape, the AIT tapes are the way to go. From what I understand, they have a chip which stores a Table Of Contents. It also has the location of each file on tape. So all the drive has to do is zip to the section on the tape, and read the file. It doesn't have to read the whole tape to find the file.

My boss went to a IBM storage demo in Vegas to see their new tech and one of the IBM guys said in a short time(meaning RSN!), they are going to have a drive that can write 40 TB to one tape! That's COOL!

Yes, RAID is not the answer

[rangek]

This is an important point I hear a lot when discussing backup solutions. "Why don't we just get a RAID array?" Or better yet, "Why don't we just back up to a big disk on another machine?"

Because, as the previous poster mentioned in passing:

RAID-5 is not an answer, because the tapes were not only in case of disk crash/user error, but also fire, flood or other catastrophe. Our backup tapes were kept at an offsite secure storage location for this reason.

If you want real security from your backup solution, it must be physically (and especially electrically) separated from your system. If your RAID array is hit by lightning, a bomb, a tornado, what-ever, you are screwed! But if your backup tapes from last week/month/whenever are in a safe deposit box somewhere, you can restore your data to your new location/machines.

So regardless of the merits of the question about tape size, tape technology is important and should be a topic of discussion among all would be geeks.

Linear Tape Open

[slkojfosijls]

have you looked at linear tape open (LTO)? Go to http://www.LTO.org. This is a consortium that IBM, HP, Seagate and others are part of that increases the capacity of a DLT-sized (approx) cartridge to 100 GB uncompressed. The most impressive thing (IMHO) is that 8 year plan which doubles capacity every 2 years or so. In 8 years a 800 GB uncompressed cartridge. With a ADSM/TSM type solution you can hopefully keep that cart pretty full.

Re:What about tape changers?

[kenn]

Be careful with DDS tape changers. DDS driver are rated at a relatively low duty cycle (about 20%), so the theoritical capacity might not be a large as you think.

20% means only about 5 hours or operation per day. Look at your throughput in MB/min and work out how much you can store in 5 hours.

DDS tape changers are meant to reduce the amount of human labour involved in rotating tapes. They are not meant to be used to run DDS drives 100% of the time.

I'd say go for DLT if you are hitting the limits of DDS. DLT is rated at 100% duty cycle. You can run those babies all day and all night without worry.

Later,
Kenn

smaller tapes - easier automated handling

[phr1]

I've heard that vendors have moved toward physically smaller tapes because automated libraries are easier to build when the tapes are smaller. For example, a DAT library mechanism is quite a bit cheaper than an 8mm library, which in turn is cheaper than a DLT library, and so on.

On a related note though, the cost of backups is getting ridiculous. 5 years ago, disk space cost $250/GB and tape media (DDS-2 tape) was $5/GB, so backing up $1000 worth of disk space (4 GB) cost $20. Today, disk space is about $4/GB (75 GB Maxtor drive $300) but tape media (Exabyte Mammoth, AIT, DLT, etc.) is all well over $1/GB, so backing up $1000 worth of disks is going to cost over $250. So tape may be pricing itself out of existence.