Hierarchical aggregation in networked data management

Abstract

The recent trend towards peer-to-peer and networked data management raises some challenging issues regarding data placement and processing. Additionally, as data management environments change from a machine into a local area network and from there into a global inter-network, the context of application of parallel query processing changes. In this paper we analyze parallel processing of aggregation queries in different networked contexts. First we describe briefly the Node-Partitioned Data Manager architecture and the aggregation processing in that architecture. We identify a performance bottleneck in the basic typical parallel aggregation strategy and discuss the use of hierarchical aggregation to overcome the problem. We analyze and compare the strategies both analytically and experimentally by means of a model and a simulator capable of generating different networked settings. This allowed us to compare the influence of different parameters on the performance. We were able to show the increased efficiency of the strategy and also to analyze and obtain interesting results of its behavior in varied settings. © Springer-Verlag Berlin Heidelberg 2005.