[vxfemboy]

Resolves #172

Summary

This PR resolves Issue #172 by eliminating centralized free-list contention and cross-core MESI cache-line invalidation in AF_XDP kernel-bypass networking. It introduces per-core UmemSlab partitioning, lock-free batch recycling between CompletionRing and FillRing, and vectorizes ring buffer batch operations.

Changes

1. Per-Core UmemSlab Partitioning (crates/yip-io/src/af_xdp.rs):
   • Added UmemSlab managing disjoint, non-overlapping chunk partitions from a shared UmemPool mapping.
   • Added UmemPool::create_slabs and UmemPool::new_slabs to construct and partition shared memory regions without lock contention.
   • Workers allocate and recycle chunk addresses locally with zero cross-core cache invalidation or atomic locks.
   • Added slab.chunk_slice and slab.chunk_slice_mut with strict bounds checking and memory isolation guarantees.
2. Lock-Free Batch Ring Recycling:
   • Added UmemSlab::recycle_from_rings to harvest completed chunk descriptors directly from CompletionRing and push them into FillRing in batch, spilling any surplus directly back to the thread-local slab without heap allocations.
   • Added UmemSlab::populate_fill_ring for fast batch priming of fill rings.
3. Vectorized Ring Operations:
   • Replaced scalar item-by-item loops in FillRing, CompletionRing, RxRing, and TxRing with power-of-two masked contiguous slice copies (copy_from_slice), boosting circular ring batch throughput from 1,438 Mops/s to 4,057 Mops/s (0.25 ns/op).
4. Daemon Integration (bin/yipd/src/sharding.rs):
   • Multi-queue worker shards now partition a single shared UmemPool into per-worker slabs, adhering to real Linux kernel XDP_SHARED_UMEM requirements.
5. Benchmarks & Unit Tests:
   • Updated crates/yip-bench/benches/af_xdp_scale.rs to benchmark UmemSlab alloc/free, ring-to-ring recycling, and single-pool multi-slab scaling across 1, 2, 4, and 8 worker threads.
   • Added 6 comprehensive unit tests in crates/yip-io/tests/umem_test.rs validating slab allocation, bounds, batching, recycling, ring wrap-around, and Send/Sync invariants.
6. Documentation & Roadmap:
   • Updated CHANGELOG.md, ROADMAP.md, README.md, docs/03-architecture.md, and docs/05-benchmarking-and-performance.md.

Benchmark Results (af_xdp_scale)

Metric / Scenario	Baseline	UmemSlab + Vectorized Rings	Improvement
Circular Rings (Batch 32)	1,438 Mops/s (0.70 ns/op)	4,057 Mops/s (0.25 ns/op)	2.8× faster (-64.3% latency)
Ring-to-Ring Recycling (Batch 32)	Single-chunk loop	1,063 Mops/s (0.94 ns/op)	Direct ring transfer
8-Worker Aggregate Throughput	17.17 Gbps (1.677 Mpps)	19.60 Gbps (1.914 Mpps)	+2.43 Gbps line rate
Packet Drops / Out-of-Order	0 drops / 0 OOO	0 drops / 0 OOO	Strict FIFO ordering preserved

Verification

• cargo test --workspace: All 387 unit + 98 integration tests pass cleanly.
• cargo clippy --workspace --all-targets -- -D warnings: Clean.
• cargo fmt --check: Clean.
• cargo shear: Clean.