Welcome to the online resources for our paper "Quantifying TPC-H Choke Points and Their Optimizations" (under review). On this page, we share instructions and the code needed for reproducing the results presented in the paper. Additionally, we provide the raw benchmark results as generated by the hyriseBenchmarkTPCH binary.

Download tpch_cps.tgz

For any questions about setting up the environment, reproducing the benchmarks, or just to discuss the paper as such, please feel free to contact us.

Hyrise can be retrieved from Github. Our Step-by-Step Guide will take you through the steps needed to set up Hyrise, starting the console, or running the TPC-H benchmark.

The paper uses a version of Hyrise that is not merged into the master yet. This version contains additional optimizations found while working on the paper. It can be checked out in the mrks/tpch_paper branch. These changes will be included into the master over the next weeks and will become part of mainline Hyrise. We will update this page accordingly. For now, only the hyriseBenchmarkTPCH binary can be built in this branch.

To execute a benchmark as presented in the paper, first generate baseline results using

./hyriseBenchmarkTPCH -s 10 -o baseline.json

Next, apply the provided patch file using `git apply patchfile.diff` in the repositories root folder. After recompiling, re-run the benchmarks and store the results in a different file. You can then compare two runs using the `./scripts/compare_benchmarks.py` script.

• We crawled Google Scholar using this script from Strobel and Hofmann
• Raw data is contained in the downloadable repository.

The performance breakdown of Hyrise operators can be obtained by running the benchmark with the --visualize parameter and calling the `plot_performance_breakdown.py` script.

For re-running the Hyrise TPC-H Comparison shown in Figure 3, execute hyriseBenchmarkTPCH. The average time for each benchmark query is given on the console.

Instructions for executing TPC-H on MonetDB can be found here. We used MonetDB 5 server 11.35.9, which was obtained using Ubuntu's package manager.

DuckDB was built from source (70c20f28f). The patch for running it with SF 10 is part of the archive.

Please find the mentioned diff files and the raw results in the downloadable package linked above.

4.1 Join Ordering

Baseline: original_join_order.diff
(1) DPccp: unmodified Hyrise code
(2) Greedy: greedy.diff

4.2 Predicate Placement and Ordering

Baseline: no_pushdown_no_ordering.diff
(1) Pushdown without ordering: pushdown_no_ordering.diff
(2) Pushdown and ordering: pushdown_ordering.diff

4.3 Between Composition

Baseline: no_between_composition.diff
(1) Enabled composition: unmodified Hyrise code

4.4 Join-Dependent Predicate Duplication

Baseline: no_duplication.diff
(1) Enabled duplication: unmodified Hyrise code

4.5 Physical Locality

Baseline: unsorted.diff
(1) Shuffled lineitem and orders tables: shuffled.diff
(2) Physical locality without DBMS optimizations: sorted_no_pruning.diff
(3) Physical locality with DBMS optimizations but without usage of correlations: sorted_pruning_no_correlation.diff

4.6 Correlated Columns

Baseline: same as 4.5 (3)
(1) Exploitation of correlation: unmodified Hyrise code

4.7 Flattening Subqueries

Baseline: no_flattening.diff
(1) With subquery flattening: unmodified Hyrise code

4.8 Semi Join Reduction

Baseline: no_reduction.diff
(1) With indiscriminate reduction: indiscriminate_reduction.diff
(2) With cardinality-based reduction: unmodified Hyrise code

4.9 Subplan Reuse

Baseline: no_reuse.diff
(1) With subplan reuse and enabled column pruning: unmodified Hyrise code
(2) With subplan reuse and disabled column pruning: no_column_pruning.diff

5.1 Dependent Group-By Keys

Baseline: no_dependent_groupby_detection.diff
Automatic group-by detection: unmodified Hyrise code
Manual group-by detection for foreign keys: manual_groupby_detection.diff

5.2 Large IN Clauses

Baseline: no_in_reformulation.diff
Reformulation of IN clauses: unmodified Hyrise code