Document insertion into a native XML Data Store (XDS) requires to partition the document tree into a number of storage units with limited capacity, such as records on disk pages. As intra partition navigation is much faster than navigation between partitions, minimizing the number of partitions has a beneficial effect on query performance. We present a linear time algorithm to optimally partition an ordered, labeled, weighted tree such that each partition does not exceed a fixed weight limit. Whereas traditionally tree partitioning algorithms only allow child nodes to share a partition with their parent node (i.e. a partition corresponds to a subtree), our algorithm also considers partitions containing several subtrees as long as their roots are adjacent siblings. We call this sibling partitioning. Based on our study of the optimal algorithm, we further introduce two novel, near-optimal heuristics. They are easier to implement, do not need to hold the whole document instance in memory, and require much less runtime than the optimal algorithm. Finally, we provide an experimental study comparing our novel and existing algorithms. One important finding is that compared to partitioning that exclusively considers parent-child partitions, including sibling partitioning as well can decrease the total number of partitions by more than 90%, and improve query performance by more than a factor of two.
Additional information:
Das Dokument wird vom Publikationsserver der Universitätsbibliothek Mannheim bereitgestellt.
![[img]](https://madoc.bib.uni-mannheim.de/style/images/fileicons/application_pdf.png)
Search Authors in