Towards efficient main-memory use for optimum tree index update

An emerging class of database applications is characterized by frequent updates of low-dimensional data, e.g. coming from sensors that sample continuous real world phenomena. Traditional persistency requirements can be weakened in this setting of frequent updates, emphasizing a role of the main-memory in external storage index structures and enabling a higher update throughput. Moreover, in order for an index to be suitable for practical applications, efficient past-state queries should be supported without significantly penalizing other operations. These issues are not adequately addressed in the database research. We report on the R-tree—our first step towards resolving them. Based on this, we outline a number of concrete short-term and more abstract longer-term future research directions.