Report From The Tape Summit
Posted by
Howard Marks
March 21, 2012
This week I spent two busy days in San Francisco at the second annual Tape Summit. While some may wonder if there’s enough life left in tape to have a summit I know tape has taken a bad rap over the past few years. In this post I’ll reflect on a few key facts that came out at the summit and expose some fallacies about tape. To start Gartner never said “71% of all attempts to restore from tape fail”.
Now I’ve never believed that the corporate America would have stayed in business through the ‘80s and ‘90s if the then ubiquitous tape backup systems were really only 29% effective. I figured this was a quote from a Gartner report only available to subscribers and that the problem was bad interpretation of survey data so “71% of organizations had a problem with their tape backups when it came time to restore” became “71% of attempts fail” and since that would include failures caused by lost tapes, and systems that were never backed up, I could explain it away.
However at the cocktail hour I spoke at some length with Dave Russell, the Gartner analyst most Web references claim as the source of the 71% number. He told me that not only had he never written anything to that effect, but that he had searched the Gartner archives and no Gartner analyst had said such a thing anywhere he could find it. So the 71% failure rate is an urban myth.
ESG’s Mark Peters made a good point that over the past 30 years or so we asked disk storage to provide both capacity and performance. As the market is accepting that the idea of using flash memory to take over the performance, the door is again open to segregating the most active and less active data to different media. Tape, while it can’t provide random I/O performance, is still the least-expensive media for capacity.
Today’s tape is no slouch when it comes to speed either. An LTO-5 drive sucks in and spits out data at 140MB/s, plus the boost from whatever compression it was able to apply to your data, about the same as the latest Seagate 15K RPM drive, and up to twice as fast as a 720RPM drive. That’s half again as fast as a gigabit Ethernet link.
Think about the difference between a great NFL running back and a sprinter. Carl Lewis or Usain Bolt will beat Gail Sayers or Michael “the burner” Turner in straight race every time but they’re fast, like tape, but not able to make the quick change of direction. That’s why other than Bob Hayes and Renaldo “Skeets” Nehamaih, track stars don’t graduate to NFL careers.
If your application, like say video editing or rendering, loads several multi-gigabyte files to memory or a local disk, crunches them for a while and then writes its results, you might make up the 2 seconds or so it takes to get the first byte of the file by reading the rest of the file faster.
I learned a few interesting things about disk and tape technology too. The first made me reassess my estimate of how fast disk capacity will rise, and therefore disk’s cost per GB will fall. The disk industry has for the past 20 years or so used the same lithographic technology as the semiconductor industry to manufacture disk heads. Semiconductor vendors made huge investments to create machines that could etch ever-smaller lines on their chips and the disk guys bought the machines and build heads.
To keep disk density rising at 40% a year, by 2015 disk heads will need to have smaller features than semiconductor chips, making it unlikely that disks will continue to grow in density at that rate. Tape bits are 100,000 times larger than disk bits, meaning tape heads can follow, rather than lead, the related industries.
I also discovered that developments from the lab to market in the disk industry is about 1year to 18 months, while vendors in the tape business take more like 5 years to marketize gee-whiz technology like the 35GB tape drive that IBM and Fujifilm demonstrated last year. Disk development is technology-driven, while tape development is investment limited. You just can’t sell enough tape drives to justify spending enough on R&D to accelerate the process.
