# Measurements — the full reports behind the numbers

> Source: https://test-allsweb.allsweb.net/measurements
> Markdown for agents: https://test-allsweb.allsweb.net/measurements.md
> Publisher: AllsWeb (www.allsweb.com)

Every performance and security claim on this site comes from a round that was run, logged and written up. These are those write-ups in full: the hardware, the dataset, the conditions, the raw figures, and the claims that did not survive attribution and were withdrawn. Several rounds concluded that a change we expected to matter made no measurable difference — those are here too.

## Pages, in reading order

1. [The aaPanel benchmark, in full](https://test-allsweb.allsweb.net/measurements/against-aapanel) — markdown: https://test-allsweb.allsweb.net/measurements/against-aapanel.md
   - SixPanel 1.4 — the same test, run again on 28 September 2026 · Two bugs in our own benchmark, found before the numbers were used · Single runs lie on machines this size · Every endpoint · The rival was tuned first, and the gap did not close · Where the gap comes from · What the ratio means for a real shop · The slowest thing on all three boxes is none of the three · Related
2. [MariaDB versions, measured — four pairs, no reportable difference](https://test-allsweb.allsweb.net/measurements/database-versions) — markdown: https://test-allsweb.allsweb.net/measurements/database-versions.md
   - The short answer · The first database comparison was invalid, and the floor is what gave it away · The version pairs, measured properly · Why not MySQL — and this one is not a benchmark result · The claim this round withdrew · What "tested against" means, and what it no longer means · Memory matters more than the version · Where the real database wins came from · Things measured and deliberately not built · What this does not tell you · Related
3. [The OS and stack round — six axes, one reportable result](https://test-allsweb.allsweb.net/measurements/os-and-stack-round) — markdown: https://test-allsweb.allsweb.net/measurements/os-and-stack-round.md
   - The short answer · The six axes, and what each returned · The noise floor, which limits every other number on this page · Where each component comes from, and why that is the whole decision · The one OS comparison that survived attribution · Redis — the axis that decided something, and not on speed · nginx and the kernel — two washes, in full · What actually made the application faster · Claims from this round that were withdrawn · Debian 12, and why the native runtime refuses it by name · The support runway, which is what the recommendation actually rests on · A version string cannot answer a security question · What this round does not tell you · Related
4. [PHP versions, measured — four pairings, four washes](https://test-allsweb.allsweb.net/measurements/php-versions) — markdown: https://test-allsweb.allsweb.net/measurements/php-versions.md
   - The short answer · The four pairings · The three-way, in full · So 8.5 ≈ 8.3, and 8.4 is two to seven percent behind · Why the version does not decide anything · PHP 8.5 — the claim we had wrong, and how it was corrected · Two more claims from this round that were withdrawn · What this does not tell you · Related
5. [Processor architecture, measured — arm64 and x86_64, and no winner](https://test-allsweb.allsweb.net/measurements/cpu-architecture) — markdown: https://test-allsweb.allsweb.net/measurements/cpu-architecture.md
   - The short answer · The boxes · Processor — seconds for a fixed job, median of 5 · Memory and disk · What this round does not claim · Three wrong readings, caught on the way · Related
6. [Security and isolation, measured](https://test-allsweb.allsweb.net/measurements/security-and-isolation) — markdown: https://test-allsweb.allsweb.net/measurements/security-and-isolation.md
   - What was actually attempted, not asserted · The leak side and the attack side are different questions · What the kernel namespace closed, and what it cost · Per-project Redis, and the ACL that does not exist · What the hardening costs · The audit that produced this · The cache cannot serve one customer another's data · A backup is only proved by a restore · The AI assistant, reviewed as an attacker would (30 September 2026) · Related
7. [How the 6amMart optimisation was measured](https://test-allsweb.allsweb.net/measurements/what-we-fixed-in-6ammart) — markdown: https://test-allsweb.allsweb.net/measurements/what-we-fixed-in-6ammart.md
   - Three kinds of number, and why they stay apart · The admin sweep — all 297 pages · The acceptance rule · What was measured and deliberately not shipped · The load test, run on both codebases · September 2026 — the API round, on one machine · The defects are a separate kind of claim · What these numbers do not cover · Related
8. [What we withdrew](https://test-allsweb.allsweb.net/measurements/what-we-withdrew) — markdown: https://test-allsweb.allsweb.net/measurements/what-we-withdrew.md
   - The measurement that forced the rest · "+32 % from the distribution's own packages" · "MariaDB 11.x is 2.7× slower" · The database comparison that was thrown away entirely · "PHP 8.5 cannot run 6amMart" · "PHP 8.4 is about 8 % faster than 8.3" · "This kernel is about 10 ms slower per request" · Storage speed, from all of those rounds · Six axes swapped, nothing reportable on five · Two things withdrawn before they shipped, not after · One that was not a measurement at all · Related
9. [How we measure](https://test-allsweb.allsweb.net/measurements/how-we-measure) — markdown: https://test-allsweb.allsweb.net/measurements/how-we-measure.md
   - The rules, in short · The noise floor, and why it dictates everything else · Control rows are the resolution floor · Order matters more than repetition · When a live swap is impossible: A, B, A · Every row carries its status codes · Prove the arm, do not trust the swap · Two rate limiters, one fake speedup · What is held constant · The dataset · What is measured, and what it is not · Every claim about the application is checked in its own source · The raw data is archived, and the published tables are re-derivable · What we do not publish · Related
10. [Websites and apps, measured — six kinds on one server](https://test-allsweb.allsweb.net/measurements/websites-and-apps) — markdown: https://test-allsweb.allsweb.net/measurements/websites-and-apps.md
   - What was deployed · Time to answer, at the server · What was checked beyond the timing · What the round found first · Related