B-tree indexes for high update rates

Abstract

In some applications, data capture dominates query processing. For example, monitoring moving objects often requires more insertions and updates than queries. Data gathering using automated sensors often exhibits this imbalance. More generally, indexing streams is considered an unsolved problem.For those applications, B-tree indexes are good choices if some trade-off decisions are tilted towards optimization of updates rather than towards optimization of queries. This paper surveys some techniques that let B-trees sustain very high update rates, up to multiple orders of magnitude higher than traditional B-trees, at the expense of query processing performance. Not surprisingly, some of these techniques are reminiscent of those employed during index creation, index rebuild, etc., while other techniques are derived from well known technologies such as differential files and log-structured file systems.