BB-Tree: A main-memory index structure for multidimensional range queries

We present the BB-Tree, a fast and space-efficient index structure for processing multidimensional workloads in main memory. It uses a k-ary search tree for pruning and searching while keeping all data in leaf nodes. It linearizes the inner search tree and manages it in a cache-optimized array, with occasional re-organizations when data changes. To reduce the frequency of re-organizations, the BB-Tree introduces a novel architecture for leaf nodes, called bubble buckets, which automatically morphs between different representations based on their fill degree and are thus able to buffer large numbers of insertions in-place. We compare the BB-Tree to scanning, main-memory variants of the R*-tree, the kd-tree, and the VA-file, and the PH-tree using workloads over real and synthetic data. The BB-Tree is the fastest index for range queries up to a selectivity of 20%, and achieves an exact-match query performance similar to that of the best point access method, and is the most space-efficient index structure.