Object-relational DBMS for packet-level traffic analysis: case study on performance optimization

Analyzing Internet traffic at packet level involves generally large amounts of raw data, derived data, and results from various analysis tasks. In addition, the analysis often proceeds in an iterative manner and is done using ad-hoc methods and many specialized software tools. These facts together lead to severe management problems that we propose to address using a DBMS-based approach, called InTraBase. The challenge that we address in this paper is to have such a database system (DBS) that allows to perform analysis efficiently. Offthe-shelf DBMSs are often considered too heavy and slow for such usage because of their complex transaction management properties that are crucial for the usage that they were originally designed for. We describe in this paper the design choices for a generic DBS for packet-level traffic analysis that enable good performance and describe how we implement them in the case of the InTraBase. Furthermore, we demonstrate their importance through performance measurements on the InTraBase. These results provide valuable insights for researchers who intend to utilize a DBMS for packet-level traffic analysis.