Run-Length Encoding Markovian Streams

Markovian streams, a common class of imprecise data streams, have proved to be excellent models for uncertain, sequential data found in sensor readings. Radio Frequency Identification (RFID) information particularly lends itself to a Markovian model; a Markovian stream derived from one RFID trace represents all paths that an RFID tag traveled during an interval of time. Processing these types of streams, however, poses several challenges: the amount of disk space these streams use and the time it takes to process them. In this paper, we address both these challenges through run-length encoding (RLE) of Markovian streams. We introduce our algorithms for stream RLE compression and decompression, and study the effects on storage and timing efficiency and query error through experiments on real RFID-derived Markovian streams. The results of our tests prove not only that these streams are highly compressible, but also that the effects of this method of compression are ideal. RLE compression significantly reduces streams’ file size and marginally improves query processing time, all while managing the resulting query error.