Benchmark Environment

This section documents the hardware and system environment used to run all benchmarks in this guide.

System Overview

Component	Details
CPU	AMD Ryzen 7 5700U with Radeon Graphics (8 cores / 16 threads, up to 4.37 GHz, Zen 2)
CPU Architecture	x86_64 (supports AVX, AVX2, FMA, BMI1/2, SHA-NI, AES-NI)
Memory (RAM)	16 GB total (14 GiB usable)
Storage	WD PC SN560 NVMe SSD — 1 TB (PCIe NVMe, non-rotational)
Operating System	Ubuntu 24.04.2 LTS (Noble Numbat), 64-bit
Kernel Version	Linux 6.14.0-36-generic (PREEMPT_DYNAMIC)
PostgreSQL Version	16.10 (Ubuntu 16.10-0ubuntu0.24.04.1)
Extensions Used	biscuit , pg_trgm

---

Benchmark Configuration

Setting	Value	Notes
shared_buffers	16384 → 16 MB	Very small (default on many distros)
work_mem	4096 → 4 MB per query	Affects sort/hash operations
maintenance_work_mem	65536 → 64 MB	Used for CREATE INDEX
effective_cache_size	524288 → 512 MB	Planner estimate of OS page cache
synchronous_commit	ON	Safer but slightly slower writes
wal_level	replica	Allows logical/physical replication (default for modern PG)

---

Benchmark Methodology

• Queries were executed with \timing on.

• EXPLAIN ANALYZE is used for detailed plan inspection.

---

Reproducibility Instructions

To reproduce identical performance:

1. Restart PostgreSQL before cold-cache benchmarks:

   sudo systemctl restart postgresql
   

2. Ensure OS disk cache is cleared (optional, requires sudo):

   sync; echo 3 | sudo tee /proc/sys/vm/drop_caches
   

3. Disable power-saving CPU governors:

   sudo cpupower frequency-set -g performance
   

4. Run each query three times, discard the first warm run, take the median.

---

Notes

• Hardware differences significantly impact absolute latency numbers.

• Your Biscuit index speedups will vary depending on:

  • Disk speed

  • CPU branch prediction

  • Text pattern distribution

  • PostgreSQL configuration