HPSS
| HPSS |
The mass storage facility at Oak Ridge National Laboratory (ORNL) currently consists of tape and disk storage components, IBM servers, Linux servers, and High Performance Storage System (HPSS) software. As of April 2007 we have over 2 PB stored in over 9.4 million files.
Tape storage is provided by robotic tape libraries.
The StorageTek libraries can each hold up to 5,000 cartridges. The libraries house a total of eighteen 9840 drives (20 GB cartridges, uncompressed), two 9940A drives (60 GB cartridges, uncompressed) and sixteen 9940B drives (200 GB cartridges, uncompressed). The 9840 and 9940A drives read and write uncompressed data at 10 MB per second; the 9940B reads and writes at 30 MB per second. The beneficial feature of the 9840 tape technology is its fast seek time for small file access; these are the performance drives. The 9940 tape technology provides the ability to store a larger amount of data on each tape cartridge for more voluminous data sets; these are the capacity drives.
The design of the National Center for Computational Sciences (NCCS) facility allows data to be redundantly written on two separate floors of the facility. This provides for additional disaster protection because the two machine rooms are protected by different fire zones. User data is not only protected by writing twice, but also physically kept on media in separate robots on separate floors.
Four types of RAID (redundant arrays of independent disks) are used in the mass storage facility.
ORNL Mass Storage History
ORNL’s work in mass storage began in the early 1990s to support the Atmospheric Radiation Measurement project and to provide storage for simulation results generated on the NCCS’s Paragon supercomputers. To support those projects, ORNL acquired and ran the NSL UniTree storage management product.
In 1993 it became clear that NSL UniTree would eventually become inadequate for the projected needs of the NCCS. At that time a follow-on to NSL UniTree, known as HPSS, was being designed by IBM and a collaboration of Department of Energy (DOE) national laboratories (Sandia, Livermore, and Los Alamos). ORNL joined that collaboration and took on responsibility for the storage system management (SSM) portion of the product, for which the ORNL HPSS development team continues to be responsible.
ORNL continued with NSL UniTree production use until 1997, at which time the conversion to HPSS was completed.
Also in 1997 HPSS won an R&D 100 Award based on an entry initiated and prepared at ORNL.
In 1998 a proposal to establish a storage testbed was developed and sent to the Mathematics, Information and Computer Sciences (MICS) office of DOE; that testbed was named Probe. It was funded in fiscal year 1999 as a collaboration between ORNL and the National Energy Research Scientific Computing Center (known as NERSC).
The storage world, like the computing world, is constantly changing, and ORNL has changed with it. In late 2002 the ORNL storage group changed the tape robotics to StorageTek Powderhorn libraries with 9840 and 9940 tape technologies to position itself to effectively handle petabytes of data. Additionally, newer disk-array technologies are replacing older technologies, providing better performance and reliability, while taking advantage of the price benefit of these newer technologies. These types of modifications to the system will be ongoing over time to ensure that the ORNL storage system can handle the ever-growing storage capacity demands of its user community.