Some vendors, and more than a few of my fellow analysts, have declared tape dead, a market with steadily declining revenues. Spectra Logic has managed to weather the storm maintaining growth and profits selling increasingly bigger tape libraries to organizations like CERN and my neighbors in New Mexico, Sandia National Labs. At a press and analyst event this week, the company announced a new approach that frees tape from the tyranny of backup software.
Back when mainframes roamed the earth, operating systems and programming languages like COBOL allowed applications to read and write to tape data because disk was too expensive for many applications. In the PC and open systems era, applications read and write files, or more recently objects, from disk and applications no longer perform low-level block I/O.
Since traditional tape libraries require software developers to access their tape drives and in fact, the robotics that move tapes from idle slots to tape drives and back via SCSI commands, adding tape support to applications is a major undertaking. While a few archiving applications include tape support, in most data centers the only application that can access tape is the backup app. As an unfortunate result, organizations retain old backup tapes and call it an archive.
Spectra Logic has dubbed its new approach Deep Storage, and as the chief Pooh-Bah at DeepStorage, LLC I found it a bit distracting every time company executives said something like “Deep Storage is efficient.” It took half a second or so to realize they weren’t talking about me.
Spectra’s Deep Storage story opens with a new RESTful object storage API that Spectra’s dubbed DS3, for Deep Simple Storage Service. DS3 is an extension of the Amazon S3 API that’s becoming a standard for not just cloud storage services but also on-premise data center object storage systems. The object API allows organizations to take advantage of the low cost, and power consumption, of tape and still have their data available to applications written by mere mortals.
The first product to support DS3 is Spectra’s BlackPearl storage appliance. A BlackPearl appliance sits in front of a tape library, connecting to the tape library via SAS or Fibre Channel while accepting data via DS3 over 10Gbps Ethernet. This allows application developers to manage the contents of the tape library using the same get and put logic they use to write to any object store. While Spectra extended S3 to support tape better by adding commands to optimize for tape’s sequential, removable nature, BlackPearl will work with applications written using standard S3 syntax, though there will be a performance penalty.
While making tape data available via a RESTful interface is cool, BlackPearl also addresses most storage administrators' greatest fears about tape: migration. BlackPearl stores its data in the LTFS file system object by object. When you decide to upgrade from LTO-4 to LTO-7, the system will read the objects from LTO-4 cartridges and restack them down to a smaller number of denser LTO-7 cartridges in the background, then eject the old tapes for disposal. Since BlackPearl keeps the object catalog of where each object is stored, users can continue to access objects using the same URLs regardless of where they’ve been relocated to new cartridges.
I’m hoping that Spectra’s next step will be to integrate tape validation functionality. Spectra, and its competitor Quantum, have had automated tape verification as an optional feature of their libraries for a couple of years. These libraries periodically load and read each tape to verify the data is still accessible. BlackPearl could extend this functionality to not just notify the administrator that a tape has become unreadable but to then retrieve the additional copy of those objects stored on other tape(s) and construct another copy of the failed tape.
[Read about the players -- both good and bad -- involved in the Nirvanix collapse and lessons learned from the implosion of the cloud storage service in "The Nirvanix Failure: Villains, Heroes, and Lessons."]
Of course a tape-based system has to save customers a bunch of money to make up for the additional latency. Spectra plans to sell BlackPearl systems for 10 to14 cents per gigabyte for systems in the 2PB range. Since tapes and library slots are cheap, larger systems should be even more cost effective.
Even if we write all the data to two tapes and boosted the number of tape drives in the library to minimize contention, the total cost would be less than half that of even low-end disk solutions running ZFS on 4TB drives. Add in the fact that tapes on the shelf require a lot less space and power than disks in an array, and the economics are pretty attractive.
At Spectra Logic's event, beta customer Kevin Graham from Yahoo talked about how a periodic backup or even a HSM process leaves data exposed too long for Yahoo’s taste. Since BlackPearl presents an object API, Yahoo will be able to modify applications like Flicker to save objects to its primary storage and BlackPearl as users upload their photos, creating a continuous backup.
While sophisticated users like Yahoo can just write their applications for a new storage platform, the rest of us have to rely on our software vendors to support a new API like DS3. Spectra is doing what it can to make it easy by releasing a Hadoop client, a basic file mover to migrate folders from NAS to the DS3 object store, and SDKs for Python, Ruby, Java and C.
You have to hand it to Spectra Logic for making the effort to expand the market for tape; I can see many applications for this kind of technology. But the company has a challenge ahead of it in convincing steely eyed storage guys that tape really isn’t dead.
Disclaimer: Spectra Logic is a client of DeepStorage, LLC. and has provided gear for the DeepStorage Lab. Spectra paid my travel expenses for its product launch event.