Since 1986 - Covering the Fastest Computers in the World and the People Who Run Them

Language Flags
March 22, 2011

Spectra Takes Next Step in Tape Archive Reliability

Michael Feldman

Storage vendor Spectra Logic has added a series of data verification features to its tape library solution. The idea is to simplify archive administration and bring a new level of reliability into petascale tape storage. The capability will be built into the company’s next iteration of its BlueScale management software that will be released at the end of March.

The incorporation of disk-like data reliability has come about as tape is moving into the hard drive’s traditional territory: primary storage. But as data volumes grow, the expense and power requirements associated with disk storage becomes increasingly problematic, forcing users to offload more files to tape.

According to Molly Rector, Spectra’s vice president of marketing and product management, this shift has accelerated significantly over the last 18 months or so, as user data is growing at a rate beyond the practical limits of a disk-based system. Spectra is seeing a 45 percent annual growth rate for file data destined for online archives. That includes cloud data, digital archive data, HPC model data — ultimately any unstructured file data being moved off to tape libraries.

Especially as files systems get into the petabyte range, customers are looking to shift the majority of their data into online tape libraries, leaving only the active transactional data on the faster disks. File management frameworks like HPSS can be used to handle these really large environments and extend the native file system (like GPFS) out to tape. “You’re seeing about 90 percent of the storage sitting on tape in these environments,” says Rector.

For the larger HPC sites, this model is becoming especially common. Today NERSC (National Energy Research Scientific Computing Center) at Lawrence Berkeley Lab is using tape as their primary data store. In this case, they have a whopping 13 petabytes of data on tape with no secondary copy at all. For a number of applications, they’re writing directly to tape, without an intervening disk. The idea is to load the data directly from tape into computer memory, crunch the numbers for a day, a week, or whatever, and then spit the results back out to tape.

That’s more like the way mainframes used tape 20 years ago. Except this time around, the aggregate data volumes are much, much bigger. The petabyte-sized storage means users don’t want to have to make multiple copies of everything for the sake of reliability (tape media degrades over time due to moisture and dust). Instead, they want the tape system to behave like an enterprise disk, proactively notifying the proper authorities when the media is going bad so the files can be moved.

That’s essentially what Spectra’s new data integrity verification does. The software offers a three-level approach to data verification: PreScan, QuickScan and QuickScan. Each one can be turned on or off depending upon user preference.

The PreScan takes place before the tape is imported into the library in order to ensure that the media wasn’t damaged in shipment from the factory (about 1 or 2 percent of the tapes fall in this category says Rector). The idea is to prevent the user from writing to a bad tape — better to finds these things out before a production run.

Next comes QuickScan, which takes place whenever any data is written to tape. It verifies the data just written can be read back. This offers some assurance that the original data can be safely scrubbed from the disk (or flash drive or memory). A QuickScan takes just one to three minutes.

The final level is PostScan, which checks the entire tape to ensure all the sectors are readable. Essentially this is a way of automating the data integrity of the current archive, which, given the thousands of tapes involved in a typical production system, would be impractical to do manually. The scan will determine the degradation of the media, and if it has reached some pre-defined threshold, will notify the administrator.

A PostScan of one tape takes from two to three hours, which is going to tie up a drive for a good chunk time. As a result, Spectra’s BlueScale interface allow you configure the PostScan scheduling in a number of ways. For example, a user could choose to run a complete scan after a specified time period (every six months, a year, and so on) and at certain time of the day. It can also be configured to go fetch the oldest tape in the system whenever there is a drive available and run the complete scan at that point. Rector believes that’s the way most of their customers will use it.

Given the workflow and the size of the data stores, the PostScan is also more likely to be used at the large supercomputing sites rather than by Spectra’s commercial customers. In fact, Argonne National Lab and NASA Ames helped to spec out this particular capability. And a new tape library system just installed at NCSA had a hard requirement for this feature called out in the RFP.

Although the integrity features are more geared for the HPC crowd, other commercial users may find them useful as well, especially those who are socking away millions of files that need to be recalled at a moment’s notice. A good example is ESPN, which archives all their sports video. Whenever an athlete is in the news, they need to be able to tap into their multi-decade file repository for relevant material to show. And much like an HPC center, ESPN has a software team writing their own packages, in this case, custom analytics for their multimedia data. “If you didn’t know better, you’d think you were talking with Argonne National Lab,” says Rector.

Spectra, too, will be using the new integrity features in-house, even though by supercomputing standards their data set is tiny. “We’re going to run it on our own corporate data at Spectra,” explains Rector. “Our archive is going to be like 50 terabytes.”

The integrity features are built into BlueScale 11.3, which will become generally available on March 30. The new capability is free, but Spectra is recommending a tape library server upgrade for those sites with more than 5,000 tapes. The good news is that the heftier servers are also free to all customers under a service contract.