Skip to content

benchmark: add warmup to accessSync bench

Rodrigo Muino Tomonari requested to merge github/fork/RafaelGSS/add-warmup-accessSync into main

Following-up: https://github.com/nodejs/node/pull/49593#discussion_r1328052487.

Turns out, it does have a difference for the first run @anonrig. See:

➜  node git:(main) ✗ node benchmark/fs/bench-accessSync.js # firstRun
fs/bench-accessSync.js n=100000 type="existing": 1,522,375.115877485
fs/bench-accessSync.js n=100000 type="non-existing": 343,903.4938015694
fs/bench-accessSync.js n=100000 type="non-flat-existing": 1,375,793.3804884679
➜  node git:(main) ✗ node benchmark/fs/bench-accessSync.js
fs/bench-accessSync.js n=100000 type="existing": 1,316,668.6965309072
fs/bench-accessSync.js n=100000 type="non-existing": 330,345.1784036734
fs/bench-accessSync.js n=100000 type="non-flat-existing": 1,365,020.780120842
➜  node git:(main) ✗ node benchmark/fs/bench-accessSync.js
fs/bench-accessSync.js n=100000 type="existing": 1,348,179.7766454385
fs/bench-accessSync.js n=100000 type="non-existing": 294,858.29814902286
fs/bench-accessSync.js n=100000 type="non-flat-existing": 1,407,337.381426724

To get a similar behaviour you need to ensure the fs cache isn't affected (usually waiting a bit before rerunning the bench). With the warmup, the results are more consistent (tested in a dedicated machine too).

Merge request reports