Compression of High-dimensional Data Spaces Using Non-differential Augmented Vector Quantization

Most data-intensive applications are confronted with the problems of I/O bottleneck, poor query processing times and space requirements. Database compression alleviates this bottleneck, reduces disk space usage, improves disk access speed, speeds up query response time, reduces overall retrieval time and increases the effective I/O bandwidth. However, random access to individual tuples in a compressed database is very difficult to achieve with most of the available compression techniques. This paper reports a lossless compression technique called nondifferential augmented vector quantization. The technique is applicable to a collection of tuples and especially effective for tuples with numerous low to medium cardinality fields. In addition, the technique supports standard database operations, permits very fast random access and atomic decompression of tuples in large collections. The technique maps a database relation into a static bitmap index cached access structure. Consequently, we were able to achieve substantial savings in space by storing each database tuple as a bit value in the computer memory. Important distinguishing characteristics of our technique are that tuples can be compressed and decompressed individually rather than a full page or entire relation at a time. Furthermore, the information needed for tuple compression and decompression can reside in the memory. Possible application domains of this technique include decision support systems, statistical and life databases with low cardinality fields and possibly no text fields.