The work investigates two competing automatic index selection techniques designed to accelerate rules with inequality constraints. We present a new auto-index selection strategy that constructs a minimum cluster of B-Tree indexes that cover all searches with at most one inequality - the B-Tree SPS technique. The new technique ensures that rules with inequalities are evaluated efficiently, speeding up the evaluation time of real-world applications by up to 2.32x. The technique is also lightweight, incurring less than a 1% overhead in maximum memory usage and only a 6% overhead in compilation time. Furthermore, the technique is robust, with no practical degradation in performance for real-world applications when compared to the existing auto-index selection scheme.
- Fast Parallel Equivalence Relations
- Datalog Disassembly
- Provenance for Large-scale Datalog
- A Smart-Contract Decompiler - Gigahorse
- A Specialized B-Tree for Concurrent Datalog Evaluation
- Brie: A Specialized Trie for Concurrent Datalog
- Equivalence Relations in Soufflé
- High performance data structures in Soufflé
- Provenance in Soufflé
- PLDI Tutorial
- An efficient tunable selective points-to analysis for large codebases
- First conference tutorial
- Soufflé: On Synthesis of Program Analyzers
- Soufflé is open-source!
- On fast large-scale program analysis in Datalog