SQLite Benchmark Assessment Status
Current state after the local WooCommerce 5GB request-load diagnostics and the 20-repetition Real Chrome diagnostic. These runs are useful methodology and risk evidence, but they are not the final dedicated-Linux Core merge assessment.
Still requires dedicated Linux SUT plus separate Linux load generator.
5GB fixture, 4/16/32/64 concurrent requests, 3 repetitions per cell.
5GB fixture, 7 browser flows, 20 repetitions per flow per database.
Observed in the local WooCommerce rc3 request-load and Chrome diagnostics.
Published Evidence
| Report | Scope | Link |
|---|---|---|
| WooCommerce 5GB rc3 vs MariaDB vs stable | Concurrent requests 4, 16, 32, and 64 | Visual report |
| WooCommerce 5GB PR #378 assessment | PR branch vs rc3, MariaDB, and stable control | Visual report |
| WooCommerce 5GB Real Chrome browser | Chrome rendering, wp-admin, Site Editor, checkout, and frontend XHR | 20-repetition report |
rc3 vs MariaDB: Local Request Load
Negative latency deltas mean SQLite was faster. Positive throughput deltas mean SQLite completed more flows per second.
| Workload | Concurrent requests | p95 latency | Avg completion | Throughput | Write p95 | SQLite lock/busy |
|---|---|---|---|---|---|---|
| Read-heavy | 4 | +18.6% | +14.7% | -12.8% | n/a | 0 |
| Read-heavy | 16 | -5.5% | +25.2% | -20.3% | n/a | 0 |
| Read-heavy | 32 | -20.6% | +22.5% | -18.4% | n/a | 0 |
| Read-heavy | 64 | -20.7% | +36.6% | -27.1% | n/a | 0 |
| Balanced | 4 | -1.7% | -5.9% | +6.2% | -7.3% | 0 |
| Balanced | 16 | +1.0% | +30.7% | -23.7% | +0.7% | 0 |
| Balanced | 32 | -13.5% | +31.6% | -24.5% | -11.8% | 0 |
| Balanced | 64 | -13.6% | +49.1% | -32.6% | -11.8% | 0 |
| Write-heavy | 4 | -7.6% | -12.0% | +13.0% | -11.4% | 0 |
| Write-heavy | 16 | +10.4% | +17.8% | -15.5% | +1.5% | 0 |
| Write-heavy | 32 | +33.9% | +32.0% | -24.7% | +36.4% | 0 |
| Write-heavy | 64 | +40.0% | +48.7% | -32.3% | +41.5% | 0 |
Interesting local conclusion: SQLite rc3 avoided visible lock/busy failures, but write-heavy 32/64 concurrent-request cells still regressed materially in p95 latency, average completion time, throughput, and write latency. Read-heavy p95 can improve while average completion and throughput regress, so p95 alone is not enough to claim a win.
Real Chrome: rc3 vs MariaDB
| Flow | Chrome-ready | p95 ready | TTFB | XHR p95 |
|---|---|---|---|---|
| Woo checkout order submit | -0.2% | -1.7% | +12.7% | -23.7% |
| Woo frontend XHR burst | +2.5% | +0.9% | +12.9% | -1.3% |
| Woo product detail render | +0.4% | +0.8% | +19.8% | n/a |
| Woo shop catalog render | +0.8% | +0.6% | +11.3% | n/a |
| Woo admin orders screen | +13.9% | +17.8% | +291.4% | +15.1% |
| wp-admin post editor open | +0.0% | -0.5% | +3.8% | -1.0% |
| wp-admin site editor open | +0.1% | -0.0% | +20.9% | +22.3% |
Stable Plugin Control
The stable `v2.2.23` control is useful as a reference line but not a viable WooCommerce write-heavy result in this diagnostic: it produced lock/busy events in all workload types and flow/write-verification failures in balanced and write-heavy flows.
| Workload | Failed flows across 4/16/32/64 | SQLite lock/busy events across 4/16/32/64 |
|---|---|---|
| Read-heavy | 0 | 58, 36, 50, 90 |
| Balanced | 15, 28, 31, 37 | 206, 313, 312, 442 |
| Write-heavy | 26, 52, 61, 111 | 294, 396, 470, 771 |
Remaining Work For Full Scope
| Requirement | Status |
|---|---|
| Dedicated Linux SUT and separate Linux load generator | Missing |
| Small, medium, and large Core fixture snapshots for all database variants | Missing |
| Cache-off and object-cache-on 54-cell matrix with 3 publishable repetitions | Missing |
| Linux host, PHP-FPM, disk I/O, MySQL/MariaDB, SQLite WAL/checkpoint, and load-generator headroom evidence | Missing |
| Final report answering every `goal.md` acceptance question from hosted evidence | Missing |