Data compression in dynamic systems
Data compression in dynamic systems has several applications in the real world. Unlike the compression of static data, both data and intrinsic data patterns may change over time. A good compression in dynamic systems should either keep compression accurate for dynamic data or change its compression strategies for dynamic data patterns. In this thesis, we study both scenarios with applications from the real world. First, as an example of compression in dynamic systems with changing data, we
... ss a lossy compression in databases, called synopsis, which helps the query optimizer speed up the query process. We introduce new Haar wavelet synopsis for nonuniform accuracy and time-varying data that can be generated in near linear time and space, and updated in sublinear time. The effectiveness of our data synopsis is validated against other linear time methods by using both synthetic and real data sets. Second, as an example of compression in dynamic systems with changing data patterns, we propose a novel compression algorithm, called IPzip, which compresses IP network traffic both online and offline for efficient data transfer and storage. IPzip achieves better compression ratios by learning patterns residing in both data structures and content. We also propose a methodology to monitor over time the effectiveness of the current compression and start new pattern learning when intrinsic traffic structure changes. Finally, via trace-driven experiments on network traffic obtained from Tier-1 ISPs, we validate that IPzip achieves better performance compared to previous approaches.