gspivey · gspivey · May 21, 2026 · May 20, 2026 · May 20, 2026
diff --git a/README.md b/README.md
@@ -462,7 +462,7 @@ Each bullet below is a standalone, one-PR-sized deliverable unless noted otherwi
 - [ ] **3. `SocketAddrV6` through `UdpSocket`** — `bind` / `send_to` / `recv_from` / `connect` / `local_addr` / `peer_addr` accept and return IPv6 addresses. `set_only_v6` / `only_v6` socket option. `AddressFamily` state on the socket so the send/recv paths pick the right wire format.
 - [x] **4. IPv6 hardware offload flags** — TX: set `RTE_MBUF_F_TX_IPV6` + `RTE_MBUF_F_TX_UDP_CKSUM` with the IPv6 pseudo-header checksum in the UDP field. RX: validate IPv6 UDP checksums (honor `PKT_RX_L4_CKSUM_GOOD`). Software fallback on NICs without support. `has_tx_ipv6_cksum_offload()` accessor. *(PR [#55](https://github.com/gspivey/dpdk-stdlib-rust/pull/55), 8 tests)*
 - [x] **5. Link-local / scope IDs / solicited-node multicast MAC** — `fe80::/10` handling, `%ifindex` scope parsing, `33:33:ff:XX:XX:XX` MAC derivation from the low 24 bits of the target IPv6 address. Prereq for task 6 (NDP).
-- [ ] **6. NDP (Neighbor Discovery Protocol)** — `NdpHandler` mirroring `ArpHandler`: Neighbor Solicitation and Neighbor Advertisement message types, atomic NDP cache with fast-path lookup, auto-resolution on send, gratuitous NA on bind (parallel to our Gratuitous ARP feature), and seeding the cache from `/proc/net/ipv6_neigh` on Linux.
+- [x] **6. NDP (Neighbor Discovery Protocol)** — `NdpHandler` mirroring `ArpHandler`: Neighbor Solicitation and Neighbor Advertisement message types, atomic NDP cache with fast-path lookup, auto-resolution on send, gratuitous NA on bind (parallel to our Gratuitous ARP feature), and seeding the cache from `/proc/net/ipv6_neigh` on Linux. *(PR [#59](https://github.com/gspivey/dpdk-stdlib-rust/pull/59), 32 tests)*
 - [x] **7. ICMPv6 echo reply** — auto-respond to `ping6`, parallel to our existing IPv4 ICMP echo reply.
 - [x] **8. ICMPv6 error handling** — Destination Unreachable, Packet Too Big (with Next-Hop MTU), Time Exceeded, and Parameter Problem parsed and matched back to the originating socket. Plugs into the existing per-socket error queue (introduced for IPv4 ICMP errors) so `take_error()` works for IPv6 destinations too. *(PR [#58](https://github.com/gspivey/dpdk-stdlib-rust/pull/58), 24 tests)*
 - [ ] **9. Performance tests** — TRex PPS run at 64 / 512 / 1400B, plus the synthetic CPU-only benchmark, compared against the IPv4 baseline. Results posted to `docs/perf-test-log.md`. No PPS regression vs IPv4 required to cross off the IPv6 feature.

diff --git a/docs/perf-test-log.md b/docs/perf-test-log.md
@@ -10,6 +10,82 @@ Each entry captures the git context, test configuration, results, and analysis.
 
 ---
 
+## Run #18: NDP (Neighbor Discovery Protocol) — No Regression
+
+| Field | Value |
+|-------|-------|
+| **Date** | 2026-05-20 |
+| **Git Hash** | `1dd5643` |
+| **Branch** | `agent/ndp` |
+| **PR** | [#59](https://github.com/gspivey/dpdk-stdlib-rust/pull/59) |
+| **GH Actions Run** | [26163492881](https://github.com/gspivey/dpdk-stdlib-rust/actions/runs/26163492881) |
+| **Instance Type** | c6in.xlarge (4 vCPU, 6.25 Gbps baseline / 30 Gbps burst) |
+| **Traffic Generator** | TRex |
+
+### Changes Since Run #17
+
+1. **`1dd5643` — NDP (Neighbor Discovery Protocol) module.** New `dpdk-udp/src/ndp.rs` implementing RFC 4861 Neighbor Solicitation/Advertisement: parse/build NS and NA frames, NdpCache with atomic fast-path for single-peer steady state, NdpHandler mirroring ArpHandler, gratuitous NA on bind, and `/proc/net/ipv6_neigh` cache seeding. 32 unit tests. This is a new module — zero changes to the existing RX/TX hot path.
+
+### Results: Hardware (TRex)
+
+#### 64-byte packets
+
+| Target PPS | rust-dpdk RX | Drop | Kernel RX | Drop | native-dpdk RX | Drop |
+|-----------|-------------|------|----------|------|---------------|------|
+| 70,000 | 69,000 | 1.4% | 69,000 | 1.4% | 70,000 | 0.0% |
+| 140,000 | 139,000 | 0.7% | 139,000 | 0.7% | 140,000 | 0.0% |
+| 350,000 | 348,997 | 0.3% | 348,979 | 0.3% | 350,000 | 0.0% |
+| 700,000 | 697,788 | 0.3% | 382,398 | 45.4% | 699,686 | 0.04% |
+
+#### 512-byte packets
+
+| Target PPS | rust-dpdk RX | Drop | Kernel RX | Drop | native-dpdk RX | Drop |
+|-----------|-------------|------|----------|------|---------------|------|
+| 70,000 | 69,000 | 1.4% | 69,000 | 1.4% | 70,000 | 0.0% |
+| 140,000 | 139,000 | 0.7% | 139,000 | 0.7% | 140,000 | 0.0% |
+| 350,000 | 349,000 | 0.3% | 348,916 | 0.3% | 350,000 | 0.0% |
+| 700,000 | 696,778 | 0.5% | 431,308 | 38.4% | 694,499 | 0.8% |
+
+#### 1400-byte packets (near MTU)
+
+| Target PPS | rust-dpdk RX | Drop | Kernel RX | Drop | native-dpdk RX | Drop |
+|-----------|-------------|------|----------|------|---------------|------|
+| 70,000 | 69,000 | 1.4% | 69,000 | 1.4% | 70,000 | 0.0% |
+| 140,000 | 139,000 | 0.7% | 139,000 | 0.7% | 140,000 | 0.0% |
+| 350,000 | 349,000 | 0.3% | 348,651 | 0.4% | 350,000 | 0.0% |
+| 700,000 | 475,607 | 0.2% | 437,947 | 8.1% | 457,658 | 4.0% |
+
+#### 8500-byte packets (jumbo)
+
+| Target PPS | rust-dpdk RX | Drop | Kernel RX | Drop | native-dpdk RX | Drop |
+|-----------|-------------|------|----------|------|---------------|------|
+| 70,000 | 69,000 | 1.4% | 36,107 | 48.4% | 70,000 | 0.0% |
+| 140,000 | 76,140 | 2.7% | 77,729 | 0.8% | 76,031 | 2.9% |
+| 350,000 | 77,675 | 0.8% | 77,893 | 0.5% | 77,976 | 0.4% |
+
+#### tokio-dpdk (async compat layer)
+
+| Target PPS | tokio-dpdk RX | Drop |
+|-----------|--------------|------|
+| 70,000 | 69,000 | 1.4% |
+| 140,000 | 139,000 | 0.7% |
+| 350,000 | 306,839 | 12.3% |
+| 700,000 | 307,556 | 56.1% |
+
+### Analysis
+
+**No performance regression from NDP module.** NDP is a standalone new module (`ndp.rs`) that adds no code to the existing UDP RX/TX hot path. The benchmark results confirm zero measurable impact:
+
+- **rust-dpdk at 700K PPS, 64B**: 697,788 RX (0.3% drop) — matches Run #15's 665K and exceeds it, within normal variance
+- **rust-dpdk at 700K PPS, 512B**: 696,778 RX (0.5% drop) — consistent with Run #15's 657K
+- **rust-dpdk at 700K PPS, 1400B**: 475,607 RX (0.2% drop) — matches Run #15's 470K
+
+**rust-dpdk vs native-dpdk parity**: At 700K PPS with 64B packets, Rust delivers 697,788 vs native C's 699,686 — within 0.3%. At 1400B near-MTU, Rust actually beats native C (475,607 vs 457,658) due to the line-rate cap.
+
+**tokio-dpdk**: Caps at ~307K PPS at 350K+ target — improved from Run #15's ~37K cap, likely due to recent tokio compat layer improvements.
+
+---
+
 ## Run #17: ICMPv6 Error Handling — No Regression
 
 | Field | Value |

diff --git a/dpdk-udp/src/lib.rs b/dpdk-udp/src/lib.rs
@@ -58,6 +58,7 @@ pub mod geneve;
 pub mod gue;
 pub mod ipv6;
 pub mod ipv6_addr;
+pub mod ndp;
 pub mod vxlan;
 
 pub use arp::{ArpCache, ArpHandler, ArpPacket};
@@ -95,6 +96,14 @@ pub use ipv6_addr::{
     is_link_local, parse_scoped_address, solicited_node_multicast_addr,
     solicited_node_multicast_mac, ipv6_multicast_mac, ScopedIpv6Addr,
 };
+pub use ndp::{
+    NdpCache, NdpHandler, NdpPacket, ProcNeighEntry,
+    build_neighbor_solicitation, build_neighbor_advertisement, build_gratuitous_na,
+    parse_ndp_packet, parse_proc_ipv6_neigh, seed_cache_from_proc,
+    ICMPV6_TYPE_NEIGHBOR_SOLICITATION, ICMPV6_TYPE_NEIGHBOR_ADVERTISEMENT,
+    NA_FLAG_ROUTER, NA_FLAG_SOLICITED, NA_FLAG_OVERRIDE,
+    NDP_HOP_LIMIT, NDP_NS_FRAME_LEN, NDP_NA_FRAME_LEN,
+};
 
 // ============================================================================
 // Error Types