Poor performance in bench_players example.

[rj00a]

Valence Version

abf9064

What You Did

The bench_players example is indicating poor performance. This is running in release mode without any players connected:

Tick=10, MSPT=8.1462ms, Clients=0
Tick=20, MSPT=7.8771ms, Clients=0
Tick=30, MSPT=7.8221ms, Clients=0
Tick=40, MSPT=7.7786ms, Clients=0
Tick=50, MSPT=7.8106ms, Clients=0
Tick=60, MSPT=8.5027ms, Clients=0
Tick=70, MSPT=8.0414ms, Clients=0
Tick=80, MSPT=8.5672ms, Clients=0
Tick=90, MSPT=7.6703ms, Clients=0

But the idle_update benchmark is under similar conditions, and seems to suggest that a tick should take ~300 microseconds.

idle_update             time:   [378.47 µs 380.01 µs 381.72 µs]
                        change: [-5.5286% -1.2899% +3.5451%] (p = 0.63 > 0.05)

What Went Wrong

The bench_players example should have a MSPT similar to the idle_update benchmark, but it does not.

Additional Information

cargo flamegraph does not seem to indicate anything. Possibly a bug in the app's runner?

[rj00a]

Updating to bevy 0.11-dev in #392 improved the situation significantly, but there's still an unacceptably high MSPT here.

Tick=10, MSPT=2.5595ms, Clients=0
Tick=20, MSPT=2.6124ms, Clients=0
Tick=30, MSPT=2.4096ms, Clients=0
Tick=40, MSPT=2.7876ms, Clients=0
Tick=50, MSPT=2.5686ms, Clients=0
Tick=60, MSPT=2.3812ms, Clients=0
Tick=70, MSPT=2.3343ms, Clients=0
Tick=80, MSPT=2.7529ms, Clients=0
Tick=90, MSPT=2.5710ms, Clients=0
Tick=100, MSPT=1.7957ms, Clients=0
Tick=110, MSPT=2.4565ms, Clients=0
Tick=120, MSPT=0.8715ms, Clients=0

However, for some strange reason, if I have singleplayer Minecraft open on my machine at the same time I'm running bench_players, the performance actually improves.

Tick=10, MSPT=0.5308ms, Clients=0
Tick=20, MSPT=0.6093ms, Clients=0
Tick=30, MSPT=0.9646ms, Clients=0
Tick=40, MSPT=0.7730ms, Clients=0
Tick=50, MSPT=0.6068ms, Clients=0
Tick=60, MSPT=0.8253ms, Clients=0
Tick=70, MSPT=0.7890ms, Clients=0
Tick=80, MSPT=0.6498ms, Clients=0
Tick=90, MSPT=1.0591ms, Clients=0
Tick=100, MSPT=0.6587ms, Clients=0
Tick=110, MSPT=0.9885ms, Clients=0
Tick=120, MSPT=0.9323ms, Clients=0
Tick=130, MSPT=0.8012ms, Clients=0
Tick=140, MSPT=0.8606ms, Clients=0
Tick=150, MSPT=0.8374ms, Clients=0

Maybe something to do with the way the OS is scheduling the process?

I'm more inclined to believe what criterion is telling me now, but something still seems wrong here.