perf(io): tune MAP_POPULATE startup memory and optimize BatchUdpSocket burst sizing

CHANGELOG.md

@@ -29,6 +29,8 @@ until 0.1.0.
   Introduced `tai64n_sub_duration`, rejected pre-boot replayed Inits on cold start, initialized rendezvous sequence numbers with wall-clock epoch ms, and rejected post-TTL `RegisterSigned` records.
 
 ### Performance
+- **Configurable UMEM page pre-faulting (`MAP_POPULATE` vs lazy on-demand) & tunable `BatchUdpSocket` burst sizing (`crates/yip-io`, `bin/yipd`, `crates/yip-bench`, #173).**
+  Added `UmemPoolConfig` and `UmemPool::new_lazy` / `with_config` / `new_slabs_with_config` to eliminate startup latency and resident set size (RSS) bloat on memory-constrained devices. Allowing on-demand lazy paging drops UMEM initialization time from 1,590.98 µs down to **7.25 µs (219.5x faster startup)** while retaining optional page pre-faulting (`pre_fault: true`) and physical memory page locking (`libc::mlock`) for zero-jitter bare-metal environments. Generalized `BatchUdpSocket::recvmmsg_batch<const N: usize>` and `sendmmsg_burst<const N: usize>` across const-generic burst sizes, adding `BATCH_SIZE_64 = 64` and `BATCH_SIZE_128 = 128` constants to maximize single-queue syscall amortization at 40G+ line rates. Added runtime environment control in `bin/yipd` via `YIP_UMEM_LAZY` and `YIP_UMEM_MLOCK`.
 - **Per-core UMEM slab allocation & lock-free ring recycling for AF_XDP (`crates/yip-io`, `bin/yipd`, `crates/yip-bench`, #172).**
   Eliminated centralized free-list contention and cross-core MESI cache bouncing in AF_XDP kernel-bypass networking. Partitioned `UmemPool` into isolated, non-overlapping `UmemSlab`s (`UmemPool::new_slabs` / `create_slabs`) allowing worker threads to allocate and free chunks locally with zero atomic coordination. Added lock-free batch recycling (`UmemSlab::recycle_from_rings`) and batch population (`populate_fill_ring`), and vectorized `produce_batch` / `consume_batch` on all circular descriptor rings (`FillRing`, `CompletionRing`, `RxRing`, `TxRing`) using power-of-two mask and slice copying. Boosted circular ring batch throughput from 1,438 Mops/s to **4,057 Mops/s (0.25 ns/op)** and scaled 8-worker throughput to **19.60 Gbps (1.914 Mpps)** with 0 packet drops and 0 reordering.
 - **Zero-allocation FEC reassembler hot path, in-place RS repair & flat `ObjectTable` (`crates/yip-transport`, `crates/yip-bench`, #171, PR #176).**

README.md

@@ -70,7 +70,7 @@ Because throughput is inversely proportional to the square root of packet loss p
 - **Zero-Copy AF_XDP (XSK)**: Direct UMEM memory ring driver with per-core slab partitioning and lock-free ring recycling bypasses the Linux network stack, sustaining 40.27 Gbps and 3.93 Mpps across 8 CPU cores.
 - **Asynchronous `io_uring`**: High-performance multishot ring polling drives high packet rates without thread contention or syscall overhead.
 - **Multi-Queue TUN Sharding**: `IFF_MULTI_QUEUE` worker isolation maps packet flows to dedicated CPU cores via lock-free SPSC ring buffers.
-- **Vectorized Socket I/O**: Batched `recvmmsg` and `sendmmsg` with `UDP_SEGMENT` Generic Segmentation Offload (GSO) optimize standard POSIX network paths.
+- **Vectorized Socket I/O**: Batched `recvmmsg` and `sendmmsg` (tunable 32/64/128-packet bursts) with `UDP_SEGMENT` Generic Segmentation Offload (GSO) optimize standard POSIX network paths.
 - **Flat Open-Addressing Tables**: Replaces dynamic pointer maps with bounded flat slot tables using SplitMix64 fast hashing and backward-shift deletion with zero tombstones.
 
 ### 4. Post-Quantum Cryptography & Identity
@@ -206,13 +206,13 @@ Deterministic benchmarks measured across multi-region cloud WANs (AWS `us-east-1
 <!-- BENCHMARK-TABLE-START -->
 
 <!-- generated by crates/yip-bench/wan/collect.sh -->
-*Two 1-vCPU Debian 13 VPSes, ~60.939 ms apart, 2026-10-26. yip trades raw clean-path throughput for loss resilience (see the FEC row).*
+*Two 1-vCPU Debian 13 VPSes, ~54.365 ms apart, 2026-10-27. yip trades raw clean-path throughput for loss resilience (see the FEC row).*
 
 | | TCP 1-stream | UDP loss (200M) | effective loss @10% injected |
 |---|--:|--:|--:|
-| **yip** | 274.2 Mbit/s | 0.00% | 1% |
-| **WireGuard** | 451.8 Mbit/s | 0.01% | 11% |
-| **OpenVPN** | 34.0 Mbit/s | 0.04% | 10% |
+| **yip** | 253.7 Mbit/s | 0.00% | 0% |
+| **WireGuard** | 453.1 Mbit/s | 0.00% | 7% |
+| **OpenVPN** | 32.6 Mbit/s | 0.04% | 8% |
 
 <!-- BENCHMARK-TABLE-END -->
 

ROADMAP.md

@@ -14,7 +14,9 @@ maintainer.
   wire-auth 64-bit counter-bound FEC reassembly with pre-AEAD anti-replay gating, and
   bidirectional flow accounting in the traffic classifier (#11).
 - **Multi-core throughput & kernel-bypass** — multi-queue TUN worker sharding (`IFF_MULTI_QUEUE`),
-  core-pinned flow hashing, lock-free chunked nonce dispensers, circular replay windows, and
+  core-pinned flow hashing, lock-free chunked nonce dispensers, circular replay windows,
+  per-core UMEM slab partitioning with vectorized ring recycling (#172), configurable lazy on-demand paging vs `MAP_POPULATE` and `mlock` page pinning (#173),
+  tunable 32/64/128-datagram syscall burst sizing in `BatchUdpSocket` (#173), and
   AF_XDP zero-copy driver (over 40 Gbps aggregate throughput, 5.3x higher goodput than kernel
   WireGuard under 5% packet loss).
 - **Control plane & mesh routing** — multi-peer routing, self-certifying key-derived addresses