The Time-Split B-tree is a system of data retrieval which takes advantage of the best aspects of two storage mediums: cheap, sturdy, non-erasable, but slow-access optical disk storage and more expensive but faster-access magnetic disk storage. This is accomplished by placing recent versions of data on magnetic disk where it can be updated quickly and incrementally moving older data to the slower optical disk as it matures. In addition, keeping all versions of data takes the place of a log and makes recovery simple. The principal investigator plans to analyze, refine and extend the Time-Split B-tree by: parameterizing space utilization on both mediums to improve performance for an expected use pattern; providing support for several kinds of temporal semantics; comparing with other proposed methods for using both mediums; providing support for large hierarchical versioned objects; allowing intermediate versions of records to be discarded; and, consolidating current (magnetic disk) data when old versions are migrated to optical disk. The research will contribute to the understanding of the interaction between access methods, transaction processing, temporal semantics, versioning and the interplay of storage mediums with different characteristics.
