Hi,
We are running a self-hosted LanguageTool server (version 6.7) for a corporate Chrome extension used by ~200 employees. The extension checks Russian text in real-time as users type.
Our server setup:
- 4 vCPU × 3.3GHz, 8GB RAM
- Docker container: meyay/languagetool:latest
- JAVA_XMS=1g, JAVA_XMX=6g
- No n-gram models
- fastText enabled for language detection
- cacheSize=1000
- Single instance on port 8010
Current performance:
- After JVM warm-up: ~200-300ms per request 
- Cold start (first request after container restart): 6-11 seconds 
Users complain about slowness during the working day, not just after restarts. We use a debounce of 300ms in the extension + immediate check on space/punctuation.
Questions:
1. What are the best server.properties settings to maximize performance for Russian language checking with many concurrent users?
2. Does pipelineCaching / pipelinePrewarming actually help for Russian? We saw these options but couldn’t find benchmarks.
3. Is there a way to reduce response time below 200ms on this hardware without n-gram models?
4. Are there any known performance issues specific to Russian language rules that could cause slowness?
Thank you!