An experimental evaluation and investigation of waves of misery in r-trees

Waves of misery is a phenomenon where spikes many node splits occur over short periods time in tree indexes. negatively affect the performance indexes insertion-heavy workloads.Waves have been first observed context B-tree, these waves cause unpredictable index performance. In particular, search and index-update operations deteriorate when wave takes place, but more predictable between waves. This paper investigates presence or lack several R-tree variants, studies extent which impact each variant. Interestingly, although having poorer query performance, Linear Quadratic R-trees are found to be resilient than both Hilbert R*-trees. presents techniques reduce for One way eliminate force take place at regular times before nodes become full achieve deterministic The other that upon splitting node, do not split it evenly rather different utilization factors. allows leaf fill same pace. We study two new mitigate after has constructed, namely Regular Elective Splits (RES, short) Unequal Random (URS, short). Our experimental investigation highlights trade-offs introduced pros cons technique.