Cost- and Robustness-Based Query Optimization for Linked Data Fragments
| Autor: | Maribel Acosta, Lars Heling |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
Database
Computer science 010401 analytical chemistry InformationSystems_DATABASEMANAGEMENT 020207 software engineering 02 engineering and technology Linked data computer.file_format computer.software_genre Query optimization 01 natural sciences 0104 chemical sciences Metadata Query plan Robustness (computer science) Server 0202 electrical engineering electronic engineering information engineering SPARQL RDF computer |
| Zdroj: | Lecture Notes in Computer Science ISBN: 9783030624187 ISWC (1) |
| Popis: | Client-side SPARQL query processing enables evaluating queries over RDF datasets published on the Web without producing high loads on the data providers’ servers. Triple Pattern Fragment (TPF) servers provide means to publish highly available RDF data on the Web and clients to evaluate SPARQL queries over them have been proposed. For clients to devise efficient query plans that minimize both the number of requests submitted to the server as well as the overall execution time, it is key to accurately estimate join cardinalities to appropriately place physical join operators. However, collecting accurate and fine-grained statistics from remote sources is a challenging task, and clients typically rely on the metadata provided by the TPF server. Addressing this shortcoming, we propose CROP, a cost- and robust-based query optimizer to devise efficient plans combining both cost and robustness of query plans. The idea of robustness is determining the impact of join cardinality estimation errors on the cost of a query plan and to avoid plans where this impact is very high. In our experimental study, we show that our concept of robustness complements the cost model and improves the efficiency of query plans. Additionally, we show that our approach outperforms existing TPF clients in terms of overall runtime and number of requests. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|