|
|
|
     |
                                     |
Return to Views
We consider the view selection problem for XML content based routing: given a network, in which a stream of XML documents is routed and the routing decisions are taken based on results of evaluating XPath predicates on these documents, select a set of views that maximize the throughput of the network. While in view selection for relational queries the speedup comes from eliminating joins, here the speedup is obtained from gaining direct access to data values in an XML packet, without parsing that packet. The views in our context can be seen as a binary representation of the XML document, tailored for the network's workload. In this paper we define formally the view selection problem in the context of XML content based routing, and provide a practical solution for it. First, we formalize the problem; while the exact formulation is too complex to admit practical solutions, we show that it can be simplified to a manageable optimization problem, without loss in precision. Second we show that the simplified problem can be reduced to the Integer Cover problem. The Integer Cover problem is known to be NP-hard, and to admit a log n greedy approximation algorithm. Third, we show that the same greedy approximation algorithm performs much better on a class of workloads called 'hierarchical workloads', which are typical in XML stream processing. Namely, it returns an optimal solution for hierarchical workloads, and degrades gracefully to the log n general bound as the workload becomes less hierarchical.
@inproceedings {DBLP:conf/pods/GuptaSH03, author = {Ashish Kumar Gupta and Dan Suciu and Alon Y. Halevy}, booktitle = {PODS}, title = {The view selection problem for XML content based routing.}, pages = {68-77}, year = {2003}, url = {db/conf/pods/pods2003.html#GuptaSH03}, ee = {http://doi.acm.org/10.1145/773153.773161}, crossref = {conf/pods/2003}, bibsource = {DBLP, http://dblp.uni-trier.de} }  ©2004 Association for Computing Machinery |
