J⁺-Tree: A New Index Structure in Main Memory

Abstract

As the memory capacity increases and the hardware becomes cheaper, main memory databases (MMDB) have come true and been used in more and more applications, because they can provide better response time and throughputs. The advent of MMDB requires a reconsideration of data structures and algorithms of traditional DBMS. The index structure is one of the most important aspects that need be redesigned since it can affect the overall system performance heavily. Even though the T-tree index, which was proposed for main memory databases, has been widely accepted as a promising index structure. B⁺-tree and its variants still have their advantages in memory and are also regarded as the potential main memory database index structures. In this paper, we propose a new indexing technique called J⁺-tree for MMDB, inspired by the Judy structure which is an associative array data structure. Our J⁺-tree index not only holds the advantages of Judy (such as good single value search characteristic) but also outperforms it in many ways. For example, J⁺-tree can obtain better performance for range queries that are very slow in Judy structure. We compare the J⁺-tree index with Judy, T-tree and B⁺-tree on time and space aspects, and the experimental results show that J⁺-tree can provide better overall performance in main memory.

This work is partly supported by a grant from HP Lab China, NSFC China No.60573092, NSFC China No.60496325, China Grid No.CNGI-04-15-7A and NSFC China No.60503038.