What is Query Engine

Since the size of the underlying data warehouse server (DWS) is usually very large, response time needed for computing queries is the main issue in decision support systems (DSS). Business analysis is the main application field in the context of DSS, as well as OLAP queries being the most useful ones; in fact, these queries allow us to support different kinds of analysis based on a multi-resolution and a multi-dimensional view of the data. By performing OLAP queries, business analysts can efficiently extract summarized knowledge, by means of SQL aggregation operators, from very large repositories of data like those stored in massive DWSs. Then, the extracted knowledge is exploited to support decisions in strategic fields of the target business, thus efficiently taking advantage from the amenity of exploring and mining massive data via OLAP technologies. The negative aspect of such an approach is just represented by the size of the data, which is enormous, currently being tera-bytes and peta-bytes the typical orders of data magnitude for enterprise DWSs, and, as a consequence, data processing costs are explosive. Despite the complexity and the resource-intensiveness of processing OLAP queries against massive DWSs, client-side systems performing OLAP and data mining, the most common application interfaces versus DWSs, are often characterized by small amount of memory, small computational capability, and customized tools with interactive, graphical user interface supporting qualitative, trend analysis. For instance, consider the context of retail systems. Here, managers and analysts are very often more interested in the product-sale plot in a fixed time window rather than to know the sale of a particular product in a particular day of the year. In others words, managers and analysts are more interested in the trend analysis rather than in the punctual, quantitative analysis, which is, indeed, more proper for OLTP systems. This consideration makes it more convent and efficient to compute approximate answers rather than exact answers. In fact, typical decision-support queries can be very resource intensive in terms of spatial and temporal computational needs. Obviously, the other issue that must be faced is the accuracy of the answers, as providing fast and totally wrong answers is deleterious. All considering, the key is proving fast, exploratory answers with some guarantees on their degree of approximation. On the other hand, in the last few years, DSS have become very popular: for example, sales transaction databases, call detail repositories, customer services historical data, and so forth. As a consequence, providing fast, even if approximate, answers to aggregate queries has become a tight requirement to make DSS-based applications efficient, and, thus, has been addressed in research in the vest of the so-called approximate query answering (AQA) techniques. Furthermore, in such data warehousing environments, executing multi-steps, query-processing algorithms is particularly hard because the computational cost for accessing multi-dimensional data would be enormous. Therefore, the most important issues for enabling DSS-based applications are: (1) minimizing the time complexity of query processing algorithms by decreasing the number of the needed disk I/Os, and (2) ensuring the quality of the approximate answers with respect to the exact ones by providing some guarantees on the accuracy of the approximation. Nevertheless, proposals existent in literature devote little attention to the point (2), which is indeed critical for the investigated context.