| Resumo: | It is not new that on-line analytical processing systems arose to companies to stay. They have the ability to change the most common application scenarios that decision- makers use on their everyday tasks. The large flexibility in data exploration and high performance response levels to queries these systems have make them very useful tools for exploring multidimensional data accordingly to the most diverse analysis perspectives of decision-makers. However, despite all the computational resources and techniques we have today, sometimes, it is very hard to maintain such levels of performance for all application scenarios, analytical systems, or user demands. When context conditions and application requirements change, performance losses may occur. There are a lot of strategies, techniques and mechanism that were designed and developed to avoid (or at least to attenuate) such undesirable low performance situations with the purpose to reduce especially data servers load. On-line analytical processing systems caching is one of them, designed for maintaining previous queries and serving them upon subsequent requests without having to ask the server repeatedly. In this paper, we present an on-line analytical processing systems caching technique with the ability to identify the exploration patterns of its users, i.e., what queries a user will submit during a working session, their frequency and resources involved, and to predict what data they will request in a near future, as well as the sequence of those requests. To do that in an efficient manner, we need to maintain a positive ratio between the time spent to predict and materialize the most relevant views to users, and the time that would be spent if no prediction had been done. Using association rules and Markov chains techniques, we designed a flexible manner to provide an effective caching system for on- line analytical processing systems.
|
|---|