Tape Storage for Cold Data Archive
April 8, 2015, Data Storage Innovation Conference, Santa Clara, Ca—Osamu Shimizu from Fujifilm talked about the role of tape in archive functions. The increasing volumes of data call for a new definition of cold storage and a technology that can keep up with the volume increases.
Per a UCSC study, 90 percent of all data is never accessed again after it is stored. At the same time, the volume of cold data, that which is rarely or never accessed again continues to grow over time. As a result, the archive function needs a technology that can grow with the data and only tape or optical media meet this requirement. The basic tradeoff is cost versus performance or reliability.
Surprisingly, tape technologies are increasing over time and the capital expenditures for tape equipment are growing. All the trends show that tape equipment and media are increasing due to the reliability, security, and capacity of tape.
One problem with archive technologies is that if you need the data again, you need to have a technology that provides a high transfer rate. The NCSA developed a RAID-like array of tape drives to get a high-performance storage system to handle their 380 PB of data. Metco France puts 80 percent of their data on tape and the volumes keep going up. Currently they generate 40 TB a day, or 15 PB annually. By ’18, they expect the volume to increase ten times, so they are always looking for technologies that provide cost and area efficiencies.
Google uses tape for backup of the g-mail files. They need high reliability for backup recovery along with high transfer rates and scalability for any volume. Tape provides a good total cost of operation compared to HDDs. The root causes for data loss are 45 percent hardware and 29 percent human error. Fortunately, most of the human errors are recoverable.
The reason tape is so reliable is that the underlying technology uses a read after write to verify the data storage. In addition, the drives have strong error detection and correction capabilities. The latest generation of LTO tape is likely to be viable for over 30 years, and the storage media are separate from the transport mechanism, reducing the problems with hardware failures.
When comparing the various media available for archive, tape and Blu-Ray have access times in minutes, while HDDS come up in seconds. On the other hand, the HDD has an expected lifetime of about 5-7 years, Blu-ray up to 10, and tape at least 30. In addition, the LTO specifications call for back compatibility for two-generations of media and drives. The LTO roadmap calls for increasing capacities and higher data transfer rates.
The roadmap anticipates changes in media and transport technologies to greatly increase data density and forecasts over 100 TB in a single tape by ’22. The new media will need to have chemically stable substrates and magnetic media and particle density will have to go up. The underlying technology for the tape will have to change from the horizontal metal-oxide magnetic particles to vertical nano-particles, possible in BaO(FeO3).
The density increase comes from a combination of factors. Metal oxides for recording have random distributions and oval shapes, while the BaO has a hexagonal shape that packs well. The BaO particles are smaller than the metal and can be stacked vertically rather than having to lay flat as the metal particles have to. To get the full benefits of the newer materials, the surface roughness has to be under 1 nm.
In addition to changes in recording materials, the new tapes will have smaller track widths and read head gaps. As a result, the drives for the new tapes will need better tracking position control and data recovery that will require advanced signal processing capabilities. Initial tests show that the combination of new materials is stable over temperature and humidity for over 10 years, so additional life testing is in progress.
The next generation of tape will see first deployment next year, with 220 TB per cartridge. Tests on the new version of tape and drives are already over 154 TB, so production of the larger versions is pretty well assured.
