Abstract. In this paper we present FedBench, a comprehensive benchmark suite for testing and analyzing the performance of federated query processing strategies on semantic data. The major challenge lies in the heterogeneity of semantic data use cases, where applications may face different settings at both the data and query level, such as varying data access interfaces, incomplete knowledge about data sources, availability of different statistics, and varying degrees of query expressiveness. Accounting for this heterogeneity, we present a highly flexible benchmark suite, which can be customized to accommodate a variety of use cases and compare competing approaches. We discuss design decisions, highlight the flexibility in customization, and elaborate on the choice of data and query sets. The practicability of our benchmark is demonstrated by a rigorous evaluation of various application scenarios, where we indicate both the benefits as well as limitations of the state-of-the-art federated query processing strategies for semantic data.

Abstract. Recently, processing of queries on linked data has gained attention. We identify and systematically discuss three main strategies: a bottom-up strategy that discovers new sources during query processing by following links between sources, a top-down strategy that relies on complete knowledge about the sources to select and process relevant sources, and a mixed strategy that assumes some incomplete knowledge and discovers new sources at run-time. To exploit knowledge discovered at run-time, we propose an additional step, explicitly scheduled during query processing, called correct source ranking. Additionally, we propose the adoption of stream-based query processing to deal with the unpredictable nature of data access in the distributed Linked Data environment. In experiments, we show that our implementation of the mixed strategy leads to early reporting of results and thus, more responsive query processing, while not requiring complete knowledge.

Abstract. The amount of Linked Data is increasing steadily. Optimized top-down Linked Data query processing based on complete knowledge about all sources, bottom-up processing based on run-time discovery of sources as well as a mixed strategy that combines them have been proposed. A particular problem with Linked Data processing is that the heterogeneity of the sources and access options lead to varying input latency, rendering the application of blocking join operators infeasible. Previous work partially address this by proposing a non-blocking iterator-based operator and another one based on symmetric-hash join. Here, we propose detailed cost models for these two operators to systematically compare them, and to allow for query optimization. Further, we propose a novel operator called the Symmetric Index Hash Join to address one open problem of Linked Data query processing: to query not only remote, but also local Linked Data. We perform experiments on real-world datasets to compare our approach against the iterator-based baseline, and create a synthetic dataset to more systematically analyze the impacts of the individual components captured by the proposed cost models.

No abstract

No abstract