by Elizabeth Rosenthal
Oak Ridge National Laboratory
Results from supercomputer simulations will reside in modernized tape library
Tape recorders and videocassette tapes may be considered archaic in the entertainment industry, but magnetic tape storage remains one of the most durable, secure, and cost-effective ways to archive big data generated by supercomputers.
After researchers run code on high-performance computing (HPC) systems, key results are initially stored on disks and then transferred to tape libraries for long-term storage. These systems contain tape drives that write data onto magnetic tape and hundreds or thousands of tape cartridges. Robotic arms place these cartridges in slots for safekeeping and retrieve them when users need to access data to cite experiment outcomes in academic articles or share accomplishments with colleagues.
At the Oak Ridge Leadership Computing Facility (OLCF), the Archive team in the HPC Operations Group is replacing the current tape libraries with a new library system produced by Spectra Logic Corporation. The OLCF is a US Department of Energy (DOE) Office of Science User Facility located at Oak Ridge National Laboratory (ORNL).
Although tape drives and other individual storage components are typically replaced about every 5 years, libraries can operate much longer. Still, updates are periodically required to incorporate new technologies and enhance efficiency. Most of the OLCF’s six existing systems, three of which are currently in use, have been operating for 10 years or longer, making this large-scale upgrade an infrequent and significant undertaking.
OLCF tape libraries primarily store user data accumulated from applications run on the Titan and Summit supercomputers, as well as some information from other OLCF computing resources and different user facilities.
“We work out agreements with these groups because we have a safe place to store data, and we’re doing it at scale, so we leverage that capability across the lab as best we can,” said HPC Linux Systems Engineer Jason Anderson, who leads the Archive team.
Anderson collaborates with HPC/UNIX Storage System Administrator Gregg Gawinski to administer the OLCF’s High Performance Storage System (HPSS) software used to manage petabytes of data in tape libraries. HPSS simplifies and automates processes to ensure information is both protected and accessible.
“Once the data is transferred from disk storage to the tape archive, users have assurance that we are doing everything possible to keep that data safe and secure,” he said.
The new Spectra Logic system will improve storage efficiency by consolidating data from all of the OLCF’s current data storage systems into this single library, reducing the physical footprint by 85 percent. Still, the new system will be capable of holding over 150 petabytes of data, and the compact configuration provides the team with substantial space to eventually expand the machine to a maximum capacity of 1.52 exabytes.
“The new library has just shy of 10,000 slots for tapes,” Anderson said. “We’ll be able to easily move all 70 petabytes of existing data into the new system and have plenty of room for at least a year or two before we have to start thinking about expanding.”
Instead of individual tape slots, the library has shelves of TeraPack containers—nine tape cartridges bound together in a single drawer. The TeraPacks allow for the library’s higher tape storage density in a smaller footprint than would be possible with a traditional design. This compact architecture allows the robots to travel a linear path, loading and unloading multiple tape cartridges at a time for maximum efficiency. Now, the team is conducting final testing of these capabilities.
“Once we are done preparing, the first step will be to put the library into production, which means all new data written into HPSS will be stored on tape in the new system,” Anderson said.
The team plans to begin transferring or “repacking” data from the old libraries to the new system in December, a process Anderson estimates will take about 6 months from start to finish. When all the data has been moved, the old systems will be decommissioned.
Although this tape library upgrade mostly involves installation of new hardware components, the team plans to refine software elements early next year for additional functionality. Updated versions of HPSS will further improve the efficiency of tape loading rates to minimize delays, allow users to input a queue of multiple commands at once, and implement other beneficial features.
ORNL is managed by UT-Battelle, LLC for the Department of Energy’s Office of Science, the single largest supporter of basic research for the physical sciences in the United States. DOE’s Office of Science is working to address some of the most pressing challenges of our time. For more information, please visit https://science.energy.gov.
See this post on ORNL’s website here.
Learn more about Spectra’s TFinity® ExaScale Tape Library here.