Incremental branching adaptive radix tree

Abstract

Adaptive indexing has been an area of active research in recent years. Its main concept is to create and maintain indexes adaptively and incrementally based on the incoming workload as a part of the query execution process. Database cracking is the first introduced applicable adaptive indexing paradigm. Despite the effectiveness and lightweight of database cracking, it offers slower lookups when it is compared to modern main memory index structures like adaptive radix tree (ART). ART is a recently proposed main-memory index structure which is designed to be space-efficient yet offers high performance by adapting its internal node size based on the count of keys stored in it. In this paper, we presented IBART (incremental branching adaptive radix tree), a hybrid indexing technique that takes the main concept of database cracking and applies it to ART, generating an adaptive index in terms of its creation and maintenance and also its internal nodes size. For systems of dynamic nature and unpredictable workload, IBART is proven to be more convenient than ART.