Repository navigation
feat: implement Phase 5 - raw socket backend with PacketBackend trait - #5
Merged
Merged
Conversation
Add AF_PACKET raw socket backend as a non-DPDK fallback, with an abstract PacketBackend trait enabling runtime backend selection. New modules: - backend.rs: PacketBackend trait + BackendConfig/BackendType for runtime backend selection (DPDK, AF_PACKET, AF_PACKET+MMAP, Auto) - backend_dpdk.rs: DPDK implementation of PacketBackend wrapping Port + Mempool for high-performance packet I/O - backend_raw.rs: Linux AF_PACKET raw socket backend with support for both basic send/recv and PACKET_MMAP zero-copy modes - ring_buffer.rs: PACKET_MMAP TX/RX ring buffer implementation with tpacket_v2 headers for zero-copy packet I/O Key changes to lib.rs: - Added build_udp_frame() for backend-agnostic frame construction (builds complete Ethernet frame as Vec<u8> vs mbuf-specific build_udp_packet) - Added SocketBackend enum dispatching between DPDK and generic backends - Refactored UdpSocket send/recv to use backend abstraction, reusing existing ARP/ICMP handlers and packet parsing code unchanged - Added UdpSocket::bind_with_backend() for custom backend selection - Added create_backend() factory for runtime backend initialization - All 74 dpdk-udp tests pass, all 133 workspace tests pass https://claude.ai/code/session_013wpw76iG9rx8PxU9omE8LY
Comprehensive guide for AI agents covering architecture, crate layout, design decisions, steering rules, and quick reference for key files. CLAUDE.md symlinked to AGENTS.md for Claude Code compatibility. https://claude.ai/code/session_013wpw76iG9rx8PxU9omE8LY
5 tasks done
gspivey
pushed a commit
that referenced
this pull request
Mar 25, 2026
Remove misleading claim that abstraction layer "masked" kernel performance — a fallback to kernel sockets adds overhead, it can't make things faster. Acknowledge Run #5 kernel numbers (78% drop) are an outlier vs historical ~50-58% baseline. Revert README to ~2x DPDK advantage claim. https://claude.ai/code/session_013kEK2d9wyTRWwmoAB4VBco
gspivey
pushed a commit
that referenced
this pull request
Mar 25, 2026
Run #5 (23548559577) had anomalous kernel numbers (78% drop at 64B/700K) due to a bad EC2 instance. Run #6 (23567309410) shows normal kernel performance (51% drop) consistent with Run #4. Updated both perf-test-log.md and README performance tables with the correct data. DPDK advantage is consistently ~2x at saturation, not the 3-4x that the outlier suggested. https://claude.ai/code/session_013kEK2d9wyTRWwmoAB4VBco
gspivey
added a commit
that referenced
this pull request
Jun 6, 2026
## Roadmap Item Implements roadmap item #5: `dpdk-stdlib-quic: RX queue` ## Changes - **`DpdkRxQueue`** implementing `s2n_quic_core::io::rx::Queue` trait (`for_each`, `is_empty`) - **`parse_to_rx_datagram(frame, local_addr)`** — reuses `parse_udp_packet_ref` from `dpdk-udp`, extracts TOS byte for ECN codepoint, constructs `Header { path: DpdkPathHandle, ecn }`, returns payload as `Vec<u8>` - Public re-exports of `DpdkRxQueue` and `parse_to_rx_datagram` from the crate root ## Tests Added - `valid_parse_produces_correct_datagram` — verifies payload, ECN, remote/local addresses - `wrong_dst_port_returns_none` — frames not destined for bound port are dropped - `non_ipv4_frame_returns_none` — non-IPv4 EtherType frames discarded - `truncated_frame_returns_none` — malformed/short frames discarded - `for_each_drains_all_datagrams` — Queue trait drains and resets correctly - `ecn_all_codepoints_extracted` — all 4 ECN codepoints (NotEct, Ect1, Ect0, Ce) parsed ## Tradeoffs - Payload is copied to `Vec<u8>` (accepted v1 cost per design doc — zero-copy RX is a future optimization) - Port filtering uses the port from `local_addr` parameter (matches the event loop pattern where bound port is known) --------- Co-authored-by: Agent Router <agent@agent-router.dev>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add AF_PACKET raw socket backend as a non-DPDK fallback, with an abstract
PacketBackend trait enabling runtime backend selection.
New modules:
runtime backend selection (DPDK, AF_PACKET, AF_PACKET+MMAP, Auto)
Port + Mempool for high-performance packet I/O
for both basic send/recv and PACKET_MMAP zero-copy modes
tpacket_v2 headers for zero-copy packet I/O
Key changes to lib.rs:
(builds complete Ethernet frame as Vec vs mbuf-specific build_udp_packet)
existing ARP/ICMP handlers and packet parsing code unchanged
https://claude.ai/code/session_013wpw76iG9rx8PxU9omE8LY