StatsReduce in the cloud for approximate Analytics
Michel de Rougemont · 2014
We consider a cloud as a cluster of processors holding each a large XML tree. We present a statistical representation which can be built online on each processor and allows to approximate boolean, unary and Aggregation queries. The main result of the paper shows how these statistics can be efficiently Reduced to a master node of the cloud. We obtain an approximation of the global tree structure built from the elementary trees on each processor. In this StatsReduce model, processors only exchange statistical data with their neighbours. This technique leads to the approximation of Analytics queries on the global tree structure with a quantified confidence.