Typical HEP data analysis model has two distinct data access patterns. HEP raw data claims the vast majority of the data storage space. The raw data is read only a few times by limited numbers of ``production'' processes. After the production, the resultant physics object data (DST: Data Summary Tape) amounts a few percentage of the data storage space. DST data have to be read and analyzed frequently by many collaborators world-wide.
To store and analyze the large amount of HEP raw data, large scale storage system with high speed I/O throughput between storage and CPU is required. We have been testing several types of helical scan tape drives for this category of data access[1].
For distributing the physics object data world-wide, we also have developed a hierarchical storage management (HSM) system with a file migration capability on DCE/DFS file servers[2].