Section: New Results

Large scale data distribution

Participants : Luciana Arantes [correspondent] , Rudyar Cortes, Mesaac Makpangou, Sébastien Monnet, Pierre Sens.

The proliferation of GPS-enabled devices leads to the massive generation of geotagged data sets recently known as Big Location Data. It allows users to explore and analyse data in space and time, and requires an architecture that scales with the insertions and location-temporal queries workload from thousands to millions of users. Most large scale key-value data storage solutions only provide a single one-dimensional index which does not natively support efficient multidimensional queries. In 2016, we propose GeoTrie [29], a scalable architecture built by coalescing any number of machines organized on top of a Distributed Hash Table. The key idea of our approach is to provide a distributed global index which scales with the number of nodes and provides natural load balancing for insertions and location-temporal range queries. We assess our solution using the largest public multimedia data set released by Yahoo! which includes millions of geotagged multimedia files.

We also propose ECHO [10], a novel and lightweight solution that efficiently supports range queries over a ring-like Distributed Hash Table (DHT) structure. By implementing a tree-based index structure and an effective query routing strategy, ECHO provides low-latency and low-overhead query searches by exploiting the Tabu Search principle. Load balancing is also improved reducing the traditional bottleneck problems arising in upper level nodes of tree-based index structures such as PHT. Furthermore, ECHO copes with DHT churn problems as its index exploits logical information as opposed to static reference cache approaches or replication techniques. The performance evaluation results obtained using PeerSim simulator show that ECHO achieves efficient performance compared other solutions such as the PHT strategy and its optimized version which includes a query cache.