Repository navigation
simplify topology: single rx_queues knob replaces rx_queues × workers_per_queue - #26
Merged
Merged
Conversation
…e knob The topology previously had two dimensions: rx_queues × workers_per_queue. This collapses it to one: rx_queues. Each RSS queue gets exactly 1 processing thread. When rx_queues <= 1, the socket runs in run-to-completion mode (no pipeline threads). Auto-detection: available = lcores - 1 (reserve app thread). If available <= 1 → RTC. Otherwise rx_queues = min(available, nic_max). Changes: - Remove workers_per_queue from TopologyConfig, TopologyPlan, builder API - Remove DPDK_WORKERS_PER_QUEUE env var support - Remove auto_detect_workers() helper - Simplify auto_detect_queues(): direct mapping of available lcores - Pipeline threads = rx_queues total (1 RX dispatcher + rx_queues-1 workers) - Remove --workers CLI arg from echo app - Update all tests and documentation https://claude.ai/code/session_01AX8ikLeQqd8DpPXdjAg97d
[CI] Stage: DeployInfrastructure ready.
|
[CI] Stage: SummaryAll tests PASSED. ARP seeding: kernel /proc/net/arp (automatic)
|
✅ Integration Tests Passed (Run 23520693963)Branch: Test Results
Application Logs (last 20 lines)receiver-echo-server.log sender-echo-server.log sender-test-client.log receiver-test-client-iperf.log sender-test-client-iperf.log Full Application Logs (last 200 lines each)receiver-echo-server.logsender-echo-server.logsender-test-client.logreceiver-test-client-iperf.logsender-test-client-iperf.log
|
[Perf] Stage: DeployDeploying |
[Perf] Stage: Instances Ready
|
[Perf] Stage: TRex ConfigStarting TRex configuration (MAC discovery + NIC binding)... |
[Perf] Stage: TRex Config OK
|
[Perf] Stage: TRex StartedTRex server running. Beginning benchmarks... |
[Perf] DUT ReadyDUT instance |
[Perf] Stage: Benchmark (1/5)Running |
[Perf] Benchmark Diag:
|
[Perf] Benchmark Diag:
|
[Perf] Stage: Benchmark (2/5)Running |
[Perf] Benchmark Diag:
|
[Perf] Benchmark Diag:
|
[Perf] Stage: Benchmark (3/5)Running |
[Perf] Benchmark Diag:
|
[Perf] Benchmark Diag:
|
[Perf] Stage: Benchmark (4/5)Running |
[Perf] Benchmark Diag:
|
[Perf] Benchmark Diag:
|
[Perf] Stage: Benchmark (5/5)Running |
[Perf] Benchmark Diag:
|
[Perf] Benchmark Diag:
|
[Perf] Diag: testpmd logtestpmd output (last 30 lines) |
[Perf] Stage: Results[03:52:03] INFO Generating markdown summary... Performance Test Results — c5n.2xlargeCommit: 1400B packets
512B packets
64B packets
|
gspivey
pushed a commit
that referenced
this pull request
Mar 25, 2026
…m defaults - README: replace "10-100x" claim with actual measured ~2x PPS improvement, remove multi-core pipeline section (experimental), fix stale --dpdk flag - echo app: rewrite to mirror plain-echo (65 lines vs 282), demonstrating the drop-in replacement story (only difference is the import line) - plain-echo: add signal handling and read timeout to match echo structure - perf-test-log: add Run #4 entry for topology simplification PR #26 - perf-tests: remove rust-dpdk-multicore from default configs (experimental) https://claude.ai/code/session_013kEK2d9wyTRWwmoAB4VBco
5 tasks done
gspivey
pushed a commit
that referenced
this pull request
Mar 25, 2026
PR #26 did have a full perf run (GH Actions 23522716883). Key findings: - rust-dpdk single-core matches native-dpdk at 700K/1400B (both 36% drop) - rust-dpdk-multicore 100% broken (--workers flag removed by simplification) - DPDK delivers ~2x throughput of kernel sockets at 64B/700K saturation https://claude.ai/code/session_013kEK2d9wyTRWwmoAB4VBco
gspivey
pushed a commit
that referenced
this pull request
May 29, 2026
Performance test results from GH Actions run 26633424088: - rust-dpdk 700K/64B: 695,587 RX (0.6% drop) — no regression vs Run #25 - rust-dpdk 700K/512B: 693,903 RX (0.9% drop) — no regression - tokio-dpdk caps at ~311K PPS — unchanged - native-dpdk baseline: 698,590 at 700K/64B Conclusion: IPv6 socket address support is performance-neutral.
gspivey
pushed a commit
that referenced
this pull request
Jun 2, 2026
Performance test results from GH Actions run 26815657085: - rust-dpdk at 700K PPS 64B: 689,799 RX (1.46% drop) - native-dpdk at 700K PPS 64B: 688,347 RX (1.66% drop) - No regression vs Run #26 Synthetic CPU-only IPv6 vs IPv4 comparison confirms expected ~28% TX overhead at 64B due to larger headers and mandatory UDP checksum. RX performance equivalent. Marks IPv6 roadmap tasks 3 and 9 as complete.
gspivey
pushed a commit
that referenced
this pull request
Jun 16, 2026
gspivey
added a commit
that referenced
this pull request
Jun 16, 2026
## ROADMAP Item #26: `dpdk-stdlib-tcp` Engine property tests Implements tasks 6.1, 6.2, 6.3 from `.kiro/specs/tcp-support/tasks.md`. ### Tests Added 15 property-based tests in `dpdk-stdlib-tcp/tests/engine_props.rs` covering: **Task 6.1 — Engine state machine properties:** - Property 11: Timer-driven segment generation (RTO retransmit + persist probe) - Property 12: TIME_WAIT/FIN_WAIT_2 cleanup (TCB freed after timeout) - Property 13: Resource limit enforcement (max_tcbs + backlog exceeded → RST) - Property 14: RST validation per RFC 5961 (exact seq aborts, in-window challenges, out-of-window drops) **Task 6.2 — Congestion control properties:** - Property 15: Flight-size invariant (unacked ≤ min(cwnd, rwnd)) - Property 16: Slow-start cwnd growth (MSS per ACK) - Property 17: Initial window formula (RFC 6928) - Property 18: Fast retransmit formula (NewReno: ssthresh + cwnd) - Property 22: Partial ACK in recovery (deflate cwnd, stay in recovery) - Property 25: Persist-never-aborts (indefinite probing without TimedOut) All tests use proptest with ≥100 cases per property. `cargo build && cargo test` passes with 0 failures. ### Tradeoffs - Properties 8, 9, 10, 21, 23 already exist in earlier PRs (engine_handshake_props.rs, established_props.rs, ooo_reorder_props.rs). This PR adds the remaining properties. - Tests exercise the engine directly via MockClock rather than through the full socket API, enabling deterministic timer testing. --------- 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.
Summary
workers_per_queuefromTopologyConfig,TopologyPlan, builder API, env vars, and CLIrx_queuesis now the single knob: each RSS queue gets exactly 1 processing threadrx_queues <= 1(or under stubs), no pipeline threads are spawnedrx_queues = min(lcores - 1, nic_max), RTC when ≤ 1 availableNet: -246 lines added, +139 lines (107 lines deleted). All 212 tests pass.
Test plan
cargo build— clean compilecargo test— all 212 tests pass (148 dpdk-udp, 51 dpdk, 8 dpdk-tokio, 3 dpdk-sys, 2 signal)https://claude.ai/code/session_01AX8ikLeQqd8DpPXdjAg97d