Closed MartijnR closed 1 year ago
Quick and dirty check: looks like the max improvement over the baseline is about 7 seconds (21% faster). For both 1 and 8 max processes. Oddly, the same amount of Chromium processes are shown in Activity Monitor.
Note that a browser will be launched for each 'process' set in config.json. Maybe that's the reason fewer processes are faster?
userDataDir
- small impact locally, BUT ON PRODUCTION SERVER MAY BE more SIGNIFICANTFor me locally sending 32 requests fails miserably. Sending 16 or 8 works fine. The latter actually takes a little less than half the time. So it may be smart to send small batches sequentially. I tried to find the optimum batch size. It seems to be 8 on my local Mac with 4 cores (11.9 seconds) (though a batch size of 4 is only very slightly slower per request). It has 8 threads but only 4 seem to hit. So a batch of twice the number of cores?
[^1]: for headless requests - I didn't check other requests
rebase with latest master and let OC test from branch
sh ./tools/benchmark-headless.sh
)16 commands:
baseline with max processes 1: 30.928, 32.307, 30.725 baseline with max processes 2: 33.072, 30.918, 30.104 baseline with max processes 8: 33.479, 33.456, 31.531
24 commands
baseline with max processes 1: failures baseline with max processes 2: baseline with max processes 8: failures
Interesting findings so far: