Enhancing the low-level tape layer of CERN Tape Archive software

CERN manages the largest scientific data archive in the HEP domain. The archive currently holds over 180 Petabytes, with forecasts of up to 100PB of new data added per year. Considering this numbers, the most cost-effective solution for storage in terms of capacity and maintenance is represented by magnetic tapes. The drawback of this solution is the access time which can raise to several minutes for a series of files. In an environment where very large volumes of physics data are being traded from tape to disk and vice versa, this issue becomes a serious performance bottleneck. This thesis introduces two low-level tape access optimizations: first, adding support for the Recommended Access Order (RAO), a mechanism offered by the tape hardware infrastructure to compute the file order corresponding to the minimum access time. We will present our solution for including RAO in the file retrieval operations in CERN tape archive software in order to benefit from the reduced reading time. The second optimization is replacing the usage of the SCSI tape specialized driver st with the generic SCSI driver sg. The aim of the latter one is to eliminate the undesired status caching and the unnecessary permission checking mechanism (Linux capabilities) performed by the st driver for a more straightforward and robust implementation.