Murali Mani
Worcester Polytechnic Institute

Title: Efficient XML Stream Processing: The Raindrop Approach

Abstract: We will examine key techniques used by Raindrop for processing
XQuery experessions over XML streams. XQuery can be considered as
consisting of three parts: (a) pattern retrieval via XPath expressions (b)
filtering via predicates and (c) output restructuring. There are at least
two different opportunities for performing pattern retrieval -- (a)
retrieve the patterns on the streaming XML tokens using automaton, (b)
extract several tokens into an appropriate object, say DOM object, and
perform navigation on this object. Which approach to use for retrieving
the different patterns will depend on the query/XML stream
characteristics. In Raindrop, we model pattern retrieval using automaton
and DOM-based navigation uniformly using an algebriac paradigm, and use
cost-based approaches for choosing an efficient plan.

Once we have come up with a query plan, we need to execute this plan
efficiently. In stream processing, this typically involves ensuring
minimum memory requirements. We achieve this in Raindrop by
outputting/purging data at the earliest, and thus ensuring that no data is
stored longer than required. We also study using schema knowledge to
further decrease memory requirements.

Acknowledgements:

Raindrop is partially sponsored by NSF grant, IIS 0414567.

Several people have been involved in this work: Elke Rundensteiner, Hong
Su, Jinhui Jian, Ming Li, Mingzhu Wei, Shoushen Wang, Drew Ditto, Bogomil
Tselkov, and others.