Strange Results in The Benchmarker

[AshtonKem]

The Rocket results for The Benchmarker Web Framework Tests show some results that seem to indicate some sort occasional performance problem. In particular the response time showed some very high 99th and 99.9th percentile latencies.

Results:
50th Percentile: 3.12ms
90th Percentile: 6.06ms
99th Percentile: 2247.48ms
99.9th Percentile: 6594.82ms

While Rocket isn't the only framework in this test to take multiple seconds to respond, it is the only Rust framework to exhibit this behavior. Given the extremely simple nature of the code under test, I suspect that there might be some latent but rare performance problem dogging Rocket.

It looks like the tests were updated yesterday with Rocket 0.4.1.

[jebrosen]

My guess is some known issues with performance and keep-alive in hyper 0.10 / rocket 0.4 - the benchmarking tool is likely running into these and it skews the results, both fairly and unfairly. I stand by my last comment on #1023:

From where I stand, trying to improve performance on hyper 0.10 (which is barely maintained) is not worth the effort compared to moving to and improving hyper 0.12. As such, I'm going to close this issue. #17 and #1008 are good places to watch for further progress.

[SergioBenitez]

Yeah, this is almost certainly a keep_alive issue. It would be useful to try re-running the benchmark with keep_alive = 0.

[jebrosen]

Closing due to inactivity and #1023 (comment)