btree4j: Disk-based Prefix B+-tree written in Pure Java

This software is originally developed for XBird based on Apache Xindice.

Features and Strength

Applied many improvements over the original Xindice's implementation as follows:

Implementes Prefix B+-tree in which prefixes are selected carefully to minimize their length. In prefix B+-tree, key prefixes are managed by a TRIE-like smart algorithm.

Rudolf Bayer and Karl Unterauer. "Prefix B-trees", Proc. ACM Trans. Database Syst. 2, 1, pp.11-26), March 1977. [DOI]

Pointers are compressed using Variable Byte Codes so that more keys/values are fit in memory.
Support both unique and non-unique indexing. Storing duplicate keys is allowed for non-unique indexing.
Index file based on B+-tree is supported. Multiple values per a key is also supported. BIndexFile stores <bytes[] KEY, byte[] VALUE> with VALUE stored on distinct data pages and pointers to them are managed by B+-Tree to avoid consuming many disk pages for large values in B+-tree.
Support variable-length key/value
Prefix and range search is supported in addition to exact match using a flexible callback handler. Support LIKE match using wildcards.
Paging (virtual memory) support using LRU cache replacement policy and Freespace management.
Deletion and updates are, of course, supported.
Support efficient Bulk-loading.
Minimum dependencies to external libraries. Runs on Java 7.