Low-memory JVM experiments. Graal Native. HotSpot intrinsics.
High-performance Java that defies the "enterprise bloat" stereotype. These experiments maximize throughput and minimize footprint via off-heap memory, SIMD, and native compilation.
| Module | RSS (Native) | Key Technique |
|---|---|---|
| No-GC Parser | 4.2 MB | Off-heap MemorySegment → AST |
| Native CLI | 8.0 MB | Reflection-free GraalVM Native Image |
| SIMD Math | 6.1 MB | VectorAPI + intrinsic inlining |
| Zero-Alloc Streams | 5.8 MB | Primitive-specialized Spliterators |
- Zero Reflection: Reflection-free architectures for maximum GraalVM compatibility.
- Memory Bound: Static heap targets of ≤ 32MB.
- Modern Core: Java 21+ Virtual Threads (Loom) and Structured Concurrency.
- No Bloat: Zero dependencies. No Spring, no Hibernate, no Jakarta.
| Task | Input | Java (Native) | Rust | Python |
|---|---|---|---|---|
| JSON Parse | 1MB | 3.2ms | 2.1ms | 187ms |
| CLI Startup | - | 12ms | 89μs | 245ms |
| SIMD Vector | 10^6 | 1.1ms | 0.9ms | 412ms |
Java is often misused. We treat the JVM as a systems-level tool to achieve:
- Mechanical Sympathy: Aligning data structures with L1/L2 cache lines.
- Deterministic Latency: Eliminating GC pauses via stack-allocation and off-heap storage.
- Instant Warmup: Leveraging AOT (Ahead-of-Time) compilation.
# Standard JVM (HotSpot)
./gradlew run
# Native Image (AOT)
./gradlew nativeCompile
./build/native/nativeCompile/experiment-name