Семинар за рачунарство и примењену математику, 7. мај 2019.

• 06. Мај, 2019
• Коментари (0)

Наредни састанак Семинара биће одржан у уторак, 7. маја 2019. у сали 301ф Математичког института САНУ са почетком у 14:15.

Предавач: Зоран Бабовић, Електротехнички факултет, Универзитет у Београду

Наслов предавања: NEW DISTRIBUTED IN-MEMORY RDF TRIPLE STORE IMPLEMENTATION FOR SUPPORTING ARCHITECTURE FOR SEMANTIC-BASED INTEGRATION OF SENSOR NETWORKS

Апстракт: This presentation describes architecture for semantic-based integration of sensor networks based on the distributed in-memory RDF triple store, designed with the aim to overcome the current lack of existing architectures and to enable high performance query execution in real-time interaction with semantic sensor data. The essence of this architecture is distributed RDF data management, which implies a graph-based approach, and thus several improvements have been implemented in this direction. First, the RDF triples are primarily indexed by predicates, which are also used for distribution of the triples across nodes. The selection of the primary predicate index with the accompanying index structure determined the method of maintaining the statistics of subjects and objects distribution per predicates in the RDF graph, and that is used for the estimation of the edge cardinality in the query graph and consequently for the creation of SPARQL query execution plan algorithm. These index structures are also designed for the efficient join of the sub-queries (triple patterns) results by exploiting the sort-merge join implementation approach. Furthermore, the reduction of a statically set of triples in a new added sensor observations sub-graph is performed in order to reduce the amount of processed data. In addition, new predicates are added with the goal to bypass a long join-path to sensor measurement time and value data which is utilized for optimization of time-value related queries. Also, multidimensional spatial coordinates are transformed to 1-dimensional coordinate through Z-ordering mapping and processed by special index structures for efficient execution of spatial range and kNN queries. The performance evaluation is done with sensor data from the MesoWest project, which includes data from several thounsands meteorological stations in the USA during hurricane Charley in 2004. The test results show that the implemented RDF triple store achieves from 3 to 50 times shorter query execution time from the publicly available Virtuoso RDF triple store in the conditions of performing semantic queries while intensively adding new sensor observations.