Pursuit of a scalable high performance multi-petabyte database

When the BaBar experiment at the Stanford Linear Accelerator Center starts in April 1999, it will generate approximately 200 TB/year of data at a rate of 10 MB/sec for 10 years. A mere six years later, CERN, the European Laboratory for Particle Physics, will start an experiment whose data storage requirements are two orders of magnitude larger. In both experiments, all of the data will reside in Objectivity databases accessible via the Advanced Multi-threaded Server (AMS). The quantity and rate at which the data is produced requires the use of a high performance hierarchical mass storage system in place of a standard Unix file system. Furthermore, the distributed nature of the experiment, involving scientists from 80 Institutions in 10 countries, also requires an extended security infrastructure not commonly found in standard Unix file systems. The combination of challenges that must be overcome in order to effectively deal with a multi-petabyte object oriented database is substantial. Our particular approach marries an optimized Unix file system with an industrial-strength Mass Storage System. This paper describes what we had to do to create a robust and uniform system based on these components.