Failure Resilient Aggregation in Peer-to-Peer Query Networks
Abstract
Peer-to-peer systems are used as the foundation for constructing large-scale distributed query systems. This thesis addresses the loss of accuracy in the query result experienced when nodes fail in a peer-to-peer query network. The existing peer-to-peer based aggregation systems are reviewed, and the algorithms for aggregation query execution are described and analyzed. This analysis of algorithms and properties of data loss is used when methods for increasing accuracy in aggregation queries are proposed. A simulator is constructed to investigate the behavior of the original algorithms and the modified algorithms. The simulations show the effects of node failures and how the various algorithms compare for different aggregation functions. The findings suggest that existing systems using the Hierarchical algorithm should focus on increasing the branching factor instead of replicating the aggregation process, as is suggested in existing litterature. Also, the results show that the aggregation functions react differently to some parameters, suggesting that that the aggregation algorithm should be chosen based on which aggregation functions are involved.