A Study on Log-Data Analyzing Functions in Data-Management Systems

点击翻译按钮获取中文摘要

Today' s information systems generate massive amounts of various log-data, and there are increasing needs of extracting useful knowledge from the logs and understanding what happen in the target systems. In order to satisfy these needs, this study is focused on extending existing relational database functions, and is aimed at developing efficient database-processing methods for flexible retrieval of log-data sequences, log-data transformation, and log-data analysis. Major results are as follows :Firstly, we developed a multi-dimensional data cube which supports data-mining in a multi-dimensional space of analysis. This system, named an Itemset cube, enables us to transform a given log-data set into another set of useful granules of information under a data cube model. Secondly, we developed a new efficient search algorithm, named N-OPS, for a given data sequence, by specifying a contiguous sequential pattern as a query. This pattern is a regular expression made of database predicates p(x), where x is a database record and p(x) is a SQL call with respect to x. Because N-OPS allows general database predicates, a wide range of database sequential search tasks can be supported. Our test showed that N-OPS can achieve much less times of predicate invocation than the traditional method based on NFA does. Thirdly, in order to detect a significant region in an itemset cube, we developed a method to apply Multi-Structural database operations to an itemset cube, and showed that this can automatically detect an interesting subsequence in case of computer system logs.