P99 Conf - Elevating PostgreSQL: Benchmarking Vector Search Performance - Data Set

This repo contains the associated raw benchmark results of the P99 Conf talk Elevating PostgreSQL: Benchmarking Vector Search Performance.

The results folder contains the data for the four discussed benchmark cases:

• pgvector index types
• pgvector index creation time
• pgvecto.rs performance
• pgvector vs. pgvecto.rs

For more details on the method and the benchmark objectives please refer to conference talk.

Each folder contains the data of a single run benchmark of one configuration.

---

Data Set Structure

In order to ensure full transparency and reproducibility, each data folder contains benchmark configuration data, performance data, monitoring data, cloud provider metadata, VM metadata and DBMS configuration data.

---

Benchmark Configuration Data

All configurable benchmark parameters for PostgreSQL and VectorDBBench are defined in the evaluationScenario.json.

The benchANT_versions contains the used versions of the benchANT software components to execute the benchmarks.

The execution logs of the individual benchmark steps are contained in airflowTaskInstanceDetails.json.

---

Performance Data

The raw performance data output of the VectorDBBench is contained in the 0_load.txt for the LOAD, OPTIMIZE and QUERY phase.

---

Monitoring Data

The DBMS cluster and the benchmark instances are monitored with Telegraf and the data is stored in InfluxDB.

A full snapshot of the monitoring data of each run is contained in the influx_data.zip file.

---

Cloud Provider Metadata

The cloud provider metadata for the DBMS deployment is contained in the dbms_data_resources.json / dbms_management_resources.json and for the benchmark deployment in the benchmark_resources.json.

---

VM Metadata

The VM metadata for the DBMS deployment is contained in the dbms_data_hardware_facts.json / dbms_management_hardware_facts.json and for the benchmark deployment in the benchmark_hardware_facts.json.

---

DBMS Metadata

For each DBMS, relevant configuration files and cluster states are stored before executing the workload.

DBMS-specific files are contained in each folder, e.g. postgresql.conf for PostgreSQL DBMS deployments.

---

Contact

In case of questions or feedback on the data feel free to create an issue or reach out to info@benchant.com