Hastening write operations on read-optimized out-of-core column-store databases utilizing timestamped binary association tables

Abstract

The purpose of this thesis is to extend previous research on Out-of-Core column-store databases. Following use of the Asynchronous Out-of-Core update, which kept track of data using timestamps, an appendix is created which holds the newest timestamps and updated data by appending entries to the tables as new tuples. The appendix is naturally unsorted and unindexed by nature, causing need for a linear search that is not only slow, but causes ever-increasing query time as the volume of data within the appendix expands. Although measures exist to merge the appendix with the original body of the data, which is sorted and indexed, it only makes searching on the data swifter once the merging of tuples is complete. For this reason, the use of an offset B-Tree index to allow for more efficient searches on the appendix is proposed.