Object Placement in Parallel Tape Storage Systems

High performance computing and enterprise data center require huge amount of data transfer between disk and tape. With the help of large capacity disk, writing to tape is less problematic than reading from tape. We are investigating how to use multiple tape libraries to build a parallel tape storage system with high aggregated data retrieval bandwidth. The challenge lies in that increasing data transfer parallelism may also increase tape switch time and data seek time that reduce the effective data retrieval bandwidth. We are proposing a novel scheme to place objects across tape drives based on object access pattern to reduce tape switch time and in the meanwhile increase data transfer parallelism. A multiple-tape-library simulator is built to study the proposed scheme. Simulation results show that our scheme outperforms other two previously proposed schemes with a better trade-off between tape switch time, data seek time and data transfer time