Benchmark data

The raw files behind the numbers

Every figure in the before and after table on the engineering page comes from one of these runs, and so does its description of the library they ran against. They are the benchmark harness's own output, copied here unedited: the sample timings, the counters, the app version, the capture timestamp and the machine they ran on.

First scan of the whole library

before-first-scan.json after-first-scan.json

Rescan when nothing has changed

before-sweep.json after-sweep.json

Opening the folder tree, and searching the library

before-queries.json after-queries.json

What the library is made of (archives and their contents)

corpus-first-scan.json

All 7 runs used the same read only NAS share as input and wrote only to a scratch profile, so every file here describes the same 141,696 files on the same hardware. Timings are in milliseconds. The library itself is not published: these are measurements of it, not a copy of it.

Back to how Meshory is built