CONCLUSIONS AND FUTURE TRENDS

We have discussed several approaches for supporting dynamic data cubes. The first part of the chapter mainly dealt with MOLAP data cubes and techniques that are particularly suitable for dense and low-dimensional data sets. Then techniques that explicitly address the sparseness issue were presented. The commonality between all approaches is that they try to find an appropriate balance between query, update, and storage cost. While earlier proposals mostly focused on query and storage aspects, large data sets with frequent updates created a need for more dynamic solutions.

In the future, support for sparse and high-dimensional data will become increasingly important. For example, business processes often span multiple organizational units of a company and involve multiple resources. Describing and analyzing a business process on a detailed level hence requires dealing with large numbers of attributes, i.e., high-dimensional data cubes. High dimensionality inevitably leads to sparseness since the number of data cube cells increases exponentially with the dimensionality.

Nevertheless, techniques for dense and low-dimensional data will remain important. For example, data distributions in practice are typically skewed and contain clusters with a higher density of non-empty cells. When the data cube is partitioned appropriately, some of the partitions might be dense enough to be handled like dense (sub)cubes. Also, analysts often specify selection conditions only for a subset of the dimensions. One could therefore maintain cubes for such lower-dimensional and denser projections of the data set. The Cube is another example of a structure containing dense data cubes, namely the cuboids for small numbers of grouping attributes.

Already today the size of data warehouses makes recomputation of aggregate information from scratch very costly. With increasing computing power and falling prices per storage unit, the current trend of rapidly growing data warehouses will continue and probably gain even more momentum in the near future. At the same time modern sensor technology and the ever-presence of computers in businesses result in the availability of a larger amount of information from virtually all units of a company. This poses new challenges for managing vast amounts of incoming data and fast analysis. Developing dynamic approaches for maintaining data cubes incrementally will become even more important.