The adaptive fragmentation for XML stream dissemination

The fragmentation policy over XML stream is the first major problem confronted by the XML stream dissemination system based on document fragmenting. This paper analyzes the features of the XML processing on client, network and server, and brings in the cost-model for fragmented XML stream system based on Hole-Filler model. According to the cost analysis of the parsing operation and join operation over streams on clients, this paper proposes the Adaptive XML Fragmentation algorithm (AXF) on server to dynamically merge and spit the XML fragments with XML data and query characteristics varying constantly over time. The core algorithms, A-Merge and A-Split, monitor and respond to both data and query changes automatically by using online profiler as the indication to the clients' requirements. When XML data and query characteristics were to stabilize, AXF would lead the XML fragments converge to an optimistic fragmentation rapidly. A thorough performance evaluation presents that AXF can sacrifice some fragment validity to balance the cost between transmission amount over network and the cost of query evaluation on client. And the server can achieve 2~2.5x performance improvement in scalability compared with query matching server. To the best of our knowledge, AXF is the first dynamic XML fragmenter considering merging and splitting XML fragments in an adaptive and efficient way.