Benchmark Test Environment & Disclosure
ASUS ROG Strix G16 (2023)
Performance throughput was empirically benchmarked by Om Patel on local NVMe disk with WAL durability enabled, using a synthetic 100,000-row test dataset (id INT, name VARCHAR, score DOUBLE, active BOOL, payload JSON, ~18.5 MB database footprint).
CPU
Intel Core i7-13650HX
RAM
16 GB DDR5 (4800 MT/s)
Disk I/O
1 TB Gen 5 NVMe SSD
Platform
Dart 3.12 (AOT / JIT)
Run the Reproducible Benchmark Suite
Every developer can clone the repository and verify these measurements locally on their own machines:
git clone https://github.com/ompatel3158/ULTSQL.git
cd ULTSQL
dart run tool/benchmarks/benchmark_live_comparison.dart
Throughput & Latency Breakdown
| Operation | Throughput / Latency | Storage & Features |
|---|---|---|
Public Batch API (insertBatch) |
~350,000–500,000+ rows/sec | Direct API with automatic slotted page serialization & stats |
| SQL Multi-Row Insert (Disk) | ~140,000–195,000 rows/sec | Multi-row VALUES batch with active WAL flush |
| PL/SQL Transaction Loop | ~170,000–230,000 rows/sec | In-engine JIT loop execution with B+ Tree indexing |
| Full Single-Row SQL Pipeline | ~60,000–75,000 rows/sec | Lexer, Parser, AST Planner, MVCC, and B-Tree |
| Bulk B+ Tree Index (100K Rows) | ~60–130 ms | Direct sorted batch bulk indexing |
| 768-Dim HNSW Vector Search | ~6 ms (ANN) | >99% Recall@10 verified against brute-force scan (10,000 vectors) |