Performance Engineering · Diagnostic AuditCompleted Commission

Server Profiling & 20.0 TPS Optimization

Evidence-based Spark timings investigation, heap dump analysis, and hot-path refactoring.

14.2 LaggingTPS Before

20.0 StableTPS After

18ms AvgTick Duration

The Engineering Challenge

System Bottlenecks & Constraints

A high-traffic survival network was experiencing periodic TPS dips down to 14.2 during mob farming and player trading windows. The server team had attempted random plugin removal without identifying the actual root cause.

Engineered Solution

Architectural Blueprint

Conducted deep profiling sessions using Spark timings and Java Flight Recorder (JFR) heap dumps. Traced the bottleneck to synchronous database writes inside an unoptimized inventory click listener and excessive tile entity ticking in farm chunks.

Technical Implementation Details

Hot-path listener refactoring: Shifted blocking database queries to asynchronous CompletableFuture pipelines with cached memory lookups.

Chunk ticking boundaries: Adjusted Paper configuration parameters for activation ranges and hopper transfer cooldowns.

Garbage collection tuning: Replaced default GC parameters with customized ZGC/G1GC flags optimized for high-allocation Paper workloads.

Memory leak remediation: Identified and patched unclosed HikariCP connection statements in a third-party economy bridge.

Shipped Deliverables & Scope

01Detailed Spark & JFR profiling analytical report
02Prioritized findings matrix categorized by severity and impact
03Custom patch jar resolving listener bottlenecks
04Optimized paper-world-defaults.yml and JVM startup script

Technology Stack & Tags

Spark ProfilerJava Flight RecorderVisualVMZGC Garbage CollectorPaperMC

Target PlatformPaperMC 1.21 · Spark · Java 21

Commission Similar Architecture

↗

Loading…