Max Inden
Max Inden
Next step is talking to Avast. See https://bugzilla.mozilla.org/show_bug.cgi?id=1919678#c4.
Let's wait for a benchmark run before merging here. Just to make sure nothing depends on the additional calls to `process_output`.
I suggest we land https://github.com/mozilla/neqo/pull/1929 first. Might resolve the RPS benchmark stall.
> Pulled in from `main` w/#1929. Let's see. All green. Good news 😮💨
> 1-conn/1-100mb-resp (aka. Download)/client: 💔 Performance has regressed. > > time: [1.7535 s 1.8771 s 1.9494 s] > thrpt: [51.297 MiB/s 53.273 MiB/s 57.028 MiB/s] > change: > time: [+52.443%...
Closing here. Benchmarks still show a significant performance impact when not interleaving `process_multiple_input` with `process_output`. While likely not relevant for non-localhost traffic, I don't think the small clean-up here is...
Agreed. I didn't think of `env_logger` instantiation in unit tests. Sorry about that.
Running all `quinn` unit tests takes about 5 minutes, including dependency installation, build and emulator startup.
Strange. "No change in performance detected", but a 97% increase in runtime. Will look into it now. ``` 1-conn/1-100mb-resp (aka. Download)/client: No change in performance detected. time: [358.86 ms 589.33...
Still debugging the performance regression. I expected a performance improvement. Interesting difference in `--stats` of a ~1GB download with `rx: 819060` on `main` and `rx: 25662` on `32k`. branch `main`...