youded-byte / java-optimisations-mc Goto Github PK

Hey, these flags are very interesting! But there are some issues with a few of them:

• XX:+AggressiveHeap forces ParallelGC instead of G1GC. See: https://github.com/openjdk/jdk/blob/3ebe5ad2d7771ccf7710350b749d780fef91e40c/src/hotspot/share/runtime/arguments.cpp#L1687
• XX:+EliminateLocks is already enabled by default in Java 17
• XX:JVMCIThreads=2 effectively caps the compiler thread count to 2. Java sets it to a very high value by default, and scales it up and down as needed, see: https://github.com/openjdk/jdk/blob/b38bed6d0ed6e1590a695a13a0d0c099e2bdd13a/src/hotspot/share/compiler/compileBroker.cpp#L608
• Not sure about XX:GCTimeLimit TBH, but the source describes it as this, and I'm not sure setting it to 50 is a good idea: "Limit of the proportion of time spent in GC before an OutOfMemoryError is thrown (used with GCHeapFreeLimit)" See: https://github.com/openjdk/jdk/blob/e32377e4564e636d5ebb768b9f1ead379b08f7a5/src/hotspot/share/gc/shared/gc_globals.hpp#L485

I have started a repo with a similar goal, and I will almost certainly add some of these flags there and link back here: https://github.com/brucethemoose/Minecraft-Performance-Flags-Benchmarks

I have also written a python script for objectively benchmarking Minecraft. The server benchmark times chunkgen, while the client benchmark repeats runs deterministically and records them with Intel presentmon. If you're interested, come try it out, post benches, and/or feel free to ping me about anything else!

I think that Aikar's flags should be tested in combination with these flags. I will do testing once I get my new PC, but I think it is something worth researching into.

https://aikar.co/mcflags.html