Design Efficient In-Database Video Storage Approach by Learning from Performance Evaluation of BLOB

摘要

Multimedia application designers often required to make the choice of whether storing multimedia objects as files in the file system, or as BLOBs (Binary Large Objects) in a database, or a combination of both. Towards small multimedia data, it often suggested to store them as BLOB data of database. However, previous study indicated that the efficiency of BLOB based video storage not always suffice. In this paper, By learning from the issues discovered from our previous performance evaluation, we proposed an efficient in-database video storage approach named TIViS (Temporal Interval based Video Storage). When a video object store into database by TIViS, it will be decomposed into temporal intervals and each interval will be sequential stored based on its temporal information. Additionally, in TIViS approach, a specialized buffer management mechanism is also developed to optimize the data access of multimedia objects. In our work, we implemented TIViS approach into the open source database system PostgreSQL 8.4. We conducted a series of experiments to verify the efficiency of TIViS approach, the results demonstrate that TIViS based video storage is significantly superior to traditional database system's built-in BLOB approach (e.g., PostgreSQL's Bytea).