Highly Reliable Systems: Removable Disk Backup & Recovery

Advantages of Removable Drive vs Tape

By Darren McBride
Screen Grab 3-2012

1 Bay Slimline

Many people comment they are surprised to find us still talking about tape because it’s been so long since they’ve seen it used.  It’s still used.  Extensively.  We think it’s because lot’s of IT pros are still skeptical that removable drives have a legitimate place in backup so they keep doing it the old way. Or they go for the shiny new “cloud backup” without considering how much monthly vaulting fees add up after a few years. While cloud backup can be very hands off, it’s costly and has the disadvantage that full server restores are impractical when the enterprise is down.  It just takes too long. As far as tape, the hand full of tape vendors left in the business will be rolling out their next generation tapes soon.  The marketing folks are releasing white papers and talking about their “ROI”.   Why do they always compare themselves to high end SAN, cloud, or dedupe appliances rather than Highly Reliable Systems low cost alternatives?  Let’s take a new look at this old debate.

Cost. Tape vendors are fond of discussing cost per Gigabyte of tape versus hard drive. For the last 6 months the picture hasn’t been great, but that’s partially due to Thailand flooding in 2011 driving up the price of drives.  Let’s do this honestly. Let’s look at a tape drive with similar capacity to a removable disk system for apples to apples. Tape drives are still expensive, and often require expensive SAS controllers as well.  When comparing tape to removable drive let’s talk about a small business buying a drive, controller, cables and 5 tapes for their initial rotation schedule. LTO-5 is modern tape technology with 1.5TB per tape. This size is fortuitous as we can compare 1.5TB HDD.  As shown in the adjacent screen grab the unit is being discounted online at $2450. Let’s add the vendor recommended SAS controller 462828-B21 (note: cheaper $60 SAS controllers aren’t on vendor compatibility list). We can find online at about $195 (March 2012 prices).  A cable adds another $25. Let’s suppose we’d like to purchase 5 tapes to put into a weekly rotation for the first year or so (tapes tend to wear out after 1 year despite what vendors and pundits claim).  In March 0f 2012 the price online is $51 each.  So our total is $2721 for a 5 media rotation (not including software) at discounted Internet rates.  A High-Rely slimline is $225, USB3 TI chipset PCI-e controller $55, Cable included, and 5 HR Classic 1.5TB drives are $227 each for a total of $1415 – and that’s at MSRP.

Cost Summary:
1 bay LTO tape drive, controller, and 5 1.5TB media  $2721 (discounted)
1 bay High-Rely drive, controller, and 5 1.5TB media $1415 (MSRP)

So even comparing straight retail versus discounted Internet rates, you could afford almost 6 more High-Rely removable drives for the same initial purchase price.  We could price out tape autoloaders and compare them against our 10 bay devices and show similar cost savings.  Plus, we maintain that hard drives have an average working life of 3 years whereas tape vendors essentially recommend replacing tapes yearly (see longevity below).  This fact makes the High-Rely solution even more cost effective.  See tape price compare

Longevity. Tape vendors claim their tapes can be reused 200 times (See table) but we’ve consistently seen backup and video experts use 50 as their real world guideline. I recently visited a Charter cable office and saw rows and rows of tapes on the shelf with tick marks on them.  The operator explained that they archive old commercials to tape but they discovered after 50 restores they get too many errors so they stop using them.  Very consistent with my experience in the data backup world.  Tape involves many moving parts, with dust and other damage a daily and real possibility. Normal wear and tear on tape is one of the main causes of its failing.  There is a large amount of lifespan variability in actual use. LTO uses an automatic verify-after-write technology to immediately check the data as it is being written but some backup systems explicitly perform a completely separate tape reading operation to verify the tape was written correctly. This separate verify operation doubles the number of end-to-end passes for each scheduled backup, and reduces the tape life by half.

Don’t believe us?   Certainly tape salesman and vendor spec sheets try to say otherwise.  Do this experiment: call any tape vendor’s tech support and tell them your backup is failing and that you’ve been using the same tape for a year.  We bet you’ll be told to try a new tape and call back if the problem persists.  Shelf life has often been quoted at 10 years for hard drive and up to 30 years for LTO tape, but that’s because hard drive vendors have traditionally been unwilling to provide shelf life data for fear of legal liability.

Reliability.  The failure rate for all tape backups is higher than tape vendors would like you to think.  I originally quoted some a number of articles with some shocking tape failure stats but Curtis Preston did some digging and points out many tape failure statistics out there  are more like estimates than science. Having said that, anecdotal stories like this one and this one of tape unreliability abound.  Preston did validate a statistic in a Yankee Group 2004 paper that 40.7% of 362 IT executives believed that they had suffered at least one restore failure in the previous year due to tape unreliability.  While that info is old, is based on older tape formats, and is merely an opinion survey it does illustrate the concern over tape reliability.  I’d like thank Curtis Preston for doing the hard research to debunk the bad stats out there. More honesty and hard work like his is needed in this industry.   But Curtis, when you can’t restore your data it just doesn’t matter if it’s “user error” like failing to clean the tape heads or because one of the 4 tapes you’ve spanned across gets an error.  I don’t think tape vendors bit rot statistics are reliable and the Yankee Group paper confirms my skepticism.  We find it interesting that Maxell makes a device for scanning tapes for problems (with a hard drive use the free OS utility Chkdsk).

Transportability. I don’t feel tape is the best choice for disaster recovery, or even for transport. Some industry observers claim 2.5″ drives are more transportable than 3.5″ and we’ll acknowledge that the smaller lighter drives designed for laptops have less potential for damage with physical shock.  We sell both so I have no axe to grind either way, but most customers want the sheer data capacity offered by 3.5″ drives at a lower cost.  Obviously care should be taken when transporting either drives or tape.  We also acknowledge that the higher the data density the more potential that shock could damage it.  But modern 3.5″ drives park their heads when powered down and are robust and transportable. MILLIONS of 3.5″ hard drives get shipped by UPS every day and arrive fully functional (a small percentage do arrive DOA no doubt…UPS is not known for being gentle with packages).  Even failed hard drives are shipped to Ontrack and Drive Savers all the time and those vendors proceed to recover data from them after the UPS gorillas shake the already broken drives just a little more on the way to the clean room. We’ll take our chances with higher volume hermetically sealed hard drives over tape any day.

Compatiblity. Did you know that tapes written to using Retrospect Version 6 can’t be read with Retrospect version 8?  And that’s just within a single software vendor! LTO is supposed to take care of head alignment problems of previous generations by using servo tracks but how much confidence do you have in being able to restore tapes across tape drives?  What if your LTO drive dies and you have to replace it with another brand? Here’s just one such story using LTO-3 where a tape written on Tandberg was unable to be read on an IBM drive.  Did you know LTO encrypting tape drives will use GCM for encryption/authentication, and this will allow the drives to achieve high performance, though it will be up to each individual tape drive vendor to select its specific GCM implementation?  What if vendor 1’s implementation is different than vendor 2?  Do you suppose you’ll be able to easily recover that encrypted data? “Encryption management software could be a factor in the interchange of an encrypted tape cartridge and should be taken into consideration when planning for encrypted cartridge interchange.” The number of tape vendors still in business is down to 5 if I recall correctly.  If your vendor goes out of business and your tape drive fails expect to spend time on Ebay praying you can find an old working drive with the same encryption firmware to restore those tapes.  These issues are much easier with NTFS formatted drives.

Speed & Compression. OK we’ll admit it.  LTO-5 backup rates are a bit higher than you get with 7200RPM drives during big backups.  LTO-5 is spec’d at 504 Gigabytes per hour (140Mbps).   You may see vendors quote 280 Mbps but that’s because the tape vendors cheat the calculations a bit by using an assumed compression rate of 2 to 1.  The optimistic assumption is that if the backup software, SAS controller, or driver can compress data BEFORE it is written to the drive, it reduces the amount of actual bits moved to the drive by half.  Then they get to claim double the throughput they can actually achieve natively (uncompressed).  Because they use a bigger compression buffer for LTO-6, they’ll be claiming 2.5 to 1 compression in that marketing literature, but expect most real world compression to remain around 1.7 to 1 and continue to expect that incompressible files won’t compress.  In the interest of full disclosure, I will admit that when I turn compression on to disk using shadowprotect the backup is slower whereas when I do it with LTO the backup is faster.  I suspect this discrepancy will disappear when I retest using a multi-core modern server.  LTO-5’s 504 Gigabytes per hour is a maximum theoretical rate that doesn’t account for the time to get the tape moving initially.  The interface is 6Gigabit SAS.  One reason you may not see the full theoretical data rate is that is very difficult to configure a server to perform sustained reads at this speed from a conventional server hard drive (or set of RAID drives).  An eSATA connected single High-Rely drive will probably top out at around 400 Gigabytes per hour and may be a bit slower when the hard drive gets very full (maybe 300 gigabytes per hour).  Although that’s not as fast as theoretical maximum of the LTO-5, we doubt most real world servers will see that much difference.  In addition drives have a speed advantage over tapes: tapes can’t be used by modern imaging software like Shadowprotect, Acronis True Image etc. Such software isn’t slowed down by large numbers of small files in many directories like most conventional tape software is because they do block level copies.  They also allow simple bare metal restores that dramatically increase recovery times.  Most of this technology is not available to tape.

Restore Speed (Single files).  Since tape is a sequential technology, restoring even a single file from a tape backup requires a lot of time and effort.  While tape vendors have improved this by embedding directories, it still isn’t as fast as tape.  Sitting around waiting on tapes to retrieve data is not productive.

Upgrade/Replace.  Need more than 1.5TB? Media failure?  Easy with the High-Rely product line – just buy raw SATA hard drives and upgrade them by replacing the old drives in the tray.  Need more than 1.5TB on LTO-5?  Get a forklift and replace the entire investment.

Incremental Backup.  The same imaging software is providing the ability to do incremental backups and “roll them up” into one big file (a synthetic full backup) so that it isn’t  necessary to manage individual incrementals to do a restore.  Because tape is linear, this is far more difficult to do.  Coupled with tapes higher failure rates and concern about a single bit error causing the entire thing not to restore, many IT professionals stay with full backups on tape. The paradigm of being able to create mirrored backups transparently using our Automatic Mirroring Technology along with imaging products changes the need to do full backups.

Volume Size. The ability to create a transportable backup that maximizes data on a single volume means more customers who don’t have to split large backups across multiple physical media.   LTO-6 will be out soon and vendors are already touting it.  The native capacity is 3.2TB.   Tape vendors routinely quote an “iffy” 8TB capacity based on assumed compression ratios of “2.5 to 1” in their press releases.  The higher compression rates are supposed to be achieved by using larger compression buffers.  We wonder why customers are willing to believe any of the other specs if they’re willing to distort speed and capacity with these assumptions?  High-Rely Classic solutions are shipping today (and have been for almost 2 years) at 3TB single drive capacities.  4TB will be shipping in the near future and our RAIDPacs are already at 9TB (in RAID 0 mode)

Power. Several studies point out that tape uses less power.  They quote huge power savings of tape versus disk by comparing de-duping appliances that run 24×7 to a tape sitting on the shelf.  We’d like to point out that like tape, power usage is zero with a removable disk on the shelf.

Not everyone agrees with the above.  A great blog post was written last year by Curtis Preston and I link to hit here to present the other side of the story.   Needless to say, we don’t agree.  This article goes through a similar price exercise as mine.

* The graphic shows LTO-4 longevity at 200 full uses quoted by Imation. Note the warning in red at the bottom about results varying.

Tapelife. Notice Caveats



What do you think?

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