Making `Arena.ofConfined()` Even Cheaper in JDK 28

Starting with JDK 28, the Foreign Function and Memory (FFM) API gains pooled confined arenas, letting small native-memory allocations be served from reusable per-thread pools instead of hitting the regular native allocator each time. The optimization requires no source code changes: Arena.ofConfined() transparently acquires a small pool (default four 64-byte pools per platform thread) on first allocation, carves out tiny allocations sequentially, zeroes and returns the pool on close, and falls back to normal allocation for larger or oddly aligned requests. Virtual threads borrow a pool from their carrier thread during a brief pinned handoff rather than maintaining their own cache. Benchmarks show large speedups for tiny allocations (e.g. roughly 6.8x to 18.9x for 5-20 byte allocations across Linux, macOS, and Windows), while larger allocations remain at parity since they use the regular path. Lifetime, zero-initialization, and thread-confinement guarantees of Arena.ofConfined() are unchanged, though the underlying native malloc/free correspondence may no longer be one-to-one.

Source: https://inside.java/2026/10/05/confined-pools. 8 Sync News only summarizes and links out; content copyright belongs to the authors and original sources.