MIDAS: Multi-attribute Indexing for Distributed Architecture Systems.

This work presents a pure multidimensional, indexing infrastructure for large-scale decentralized networks that operate in extremely dynamic environments where peers join, leave and fail arbitrarily. We propose a new peer-to-peer variant implementing a virtual distributed k-d tree, and develop efficient algorithms for multidimensional point and range queries. Scalability is enhanced as each peer has only partial knowledge of the network. The most prominent feature of our method, is that in expectance each peer maintains O (log n ) state and requests are resolved in O (log n ) hops with respect to the overlay size n. In addition, we provide mechanisms for handling peer failures and improving fault tolerance as well as balancing the load of peers. Finally, our work is complemented by an experimental evaluation, where MIDAS is shown to outperform existing methods in spatial as well as in higher dimensional settings.