Direct answer
chunk-system.io-threads under config/paper-global.yml handles chunk file I/O. Setting too many threads can contend for disk bandwidth without reducing tick lag.
How to check it
- Check whether slowdown occurs when entering already-explored terrain or during world saving rather than first-time generation.
- Compare disk active time/latency with CPU load and collect a spark profile around the event.
- Record the generated
chunk-system.io-threadsvalue and storage type.
What to change
chunk-system:
io-threads: -1 # paper-documented negative value means one I/O thread- Make a world backup and keep the baseline configuration.
- Only consider a change when there is evidence of a chunk disk-I/O backlog and available disk throughput.
- Restart after altering this global setting and compare measured load/save delay on the same known terrain.
Software and version notes
Compatibility and safety
Current Paper reference says values below zero select one I/O thread. Do not confuse this with worker-threads, which manages parallel generation. Defaults may differ by build.
Verify the result
Check whether existing chunks load more reliably without disk saturation, and that saving/shutdown remain healthy.