Four-Flow FIX Session Comparisons on Two DAC Rigs

Read each rig separately

These are kernel TCP, no-store canonical session comparisons: 1,000 warmup cycles, 10,000 measured cycles per run, three runs per flow and a target of 10,000 cycles/s. HP uses a Solarflare physical DAC and SCAN a Mellanox physical DAC, each between two namespaces on one machine. These are accepted Stage 5 topologies for this campaign.

p50, p99 and p99.9 are medians of the three run percentiles. Worst is the maximum of all three run extrema. Blocked flows have no latency or ratio. A passing individual flow does not admit a whole-language comparison. The archive preserves original bodies, owner counts, administration, primary/delay samples, packet-counter polling and independent cold receipts.

The engine comparison by language covers C++, C, Java, .NET, Rust and Go. This page shows the latest independently matched reports; other retained language results remain in that roster and archive.

Download every matched decision and source binding →

.NET Pure public session

Managed SDK physical results

24 admitted complete-RTT/callback cells: hp 4/4 flows, scan 4/4 flows. These use public ManagedSession and IFastHandler, with SDK encoding, session state and completed writes. The native engine library remains unloaded. Individual cells do not admit a current .NET peer comparison.

Configuration	HP collection	SCAN collection
Inherited foreground CPUs	4/4 in one generation	no complete generation; 4/4 retained
First allowed foreground CPU	4/4 in one generation	4/4 in one generation
First allowed foreground CPU, .NET 10	4/4 in one generation	4/4 in one generation

Retained good cells may come from different source generations. A failed fresh flow keeps its earlier good cell; it cannot complete the failed collection.

The explicit .NET 10 managed configuration passes all four Stage 5 flows on both rigs and has its own source/generation. Its eight cells remain separate from the original runtime. QuickFIX/n’s separately qualified parsing-only profile passes NOS and tick on HP’s first foreground CPU; its other physical flows retain report failures. The independently replayed .NET comparison below keeps those failed flows blocked and publishes metrics only for the two qualified HP flow pairs. The complete two-rig .NET comparison remains blocked. Inspect exact physical outcomes and remaining gates →

Inspect result summaries and percentiles → · Latest complete physical outcomes and failed gates → · Retained first listener failure → · Earlier first-CPU outcomes and retained mixed rate failure →

Compact source 41f8706e3354.

.NET public handler clocks

Independent Native .NET physical results

16 admitted complete-RTT/monotonic-handler cells: hp 8, scan 8. These individual cells do not establish a current .NET peer comparison. Each flavour is a separate configuration; passing cells across configurations are never merged.

Configuration	HP collection	SCAN collection
Inherited foreground CPUs	4/4 in one generation	no complete generation; 4/4 retained
First allowed foreground CPU	4/4 in one generation	4/4 in one generation

Retained good cells may come from different source generations. A failed fresh flow keeps its earlier good cell; it cannot complete the failed collection.

Inspect result summaries and percentiles → · Latest complete physical outcomes → · First-CPU outcomes and failed gates →

Compact source 41f8706e3354.

Corrected Rust clocks

New independent physical results

35 admitted callback/complete-RTT cells: hp 19, scan 16. Individual results do not admit a complete matched comparison. The separately verified first-CPU comparison below has its own current grade; retired reports remain superseded.

Inspect result summaries and percentiles → · Rust/NanoFIX physical gates → · TrueFix physical gates →

Compact source 41f8706e3354.

Rust

LibHFT Rust open / NanoFIX

All-four-flow comparison: blocked. Every metric cell is baseline / peer, in microseconds.

hp/sfc-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017
md	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017
md_nos_er	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017
tick_to_trade	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017

scan/mlx5-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017
md	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017
md_nos_er	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017
tick_to_trade	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017

Superseded: legacy Rust canonical clocks omit required parsing/extraction work. Corrected callback results use separate identities. Original archived grades and bytes are preserved. Supersession record →

The HP comparison receipts were written before the SSH transport ended; its original final marker is absent. The frozen source inventory was independently verified unchanged. The missing marker is preserved in the source record.

Complete matched report JSON → · Immutable retention seal →

Rust

LibHFT Rust open / TrueFix

All-four-flow comparison: blocked. Every metric cell is baseline / peer, in microseconds.

hp/sfc-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017
md	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017
md_nos_er	blocked	—	—	—	—	peer:achieved_rate_below_98pct; rust-canonical-clock-boundaries-20261017
tick_to_trade	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017

scan/mlx5-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017
md	blocked	—	—	—	—	peer:achieved_rate_below_98pct; rust-canonical-clock-boundaries-20261017
md_nos_er	blocked	—	—	—	—	peer:achieved_rate_below_98pct;scheduler_delay_above_interval; rust-canonical-clock-boundaries-20261017
tick_to_trade	blocked	—	—	—	—	rust-canonical-clock-boundaries-20261017

Complete matched report JSON → · Immutable retention seal →

C++

LibHFT CORE before-parser clock / NexusFIX

All-four-flow comparison: blocked. Every metric cell is baseline / peer, in microseconds.

hp/sfc-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	pass	17.413 / 20.810	24.037 / 33.254	28.640 / 38.274	101.206 / 167.740
md	pass	34.388 / 24.991	41.075 / 37.905	50.713 / 43.705	82.662 / 97.064
md_nos_er	pass	53.933 / 42.023	63.693 / 57.004	78.433 / 72.311	107.677 / 215.297
tick_to_trade	blocked	—	—	—	—	tick_receive_spans_differ

scan/mlx5-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	pass	27.905 / 30.570	45.130 / 57.708	53.565 / 64.678	121.130 / 124.935
md	pass	51.576 / 36.103	67.426 / 62.419	76.997 / 68.780	118.093 / 167.502
md_nos_er	pass	82.316 / 62.158	108.132 / 97.252	121.950 / 149.359	162.812 / 187.832
tick_to_trade	blocked	—	—	—	—	tick_receive_spans_differ

Complete matched report JSON → · Immutable retention seal →

C++

LibHFT CORE callback clock / NexusFIX

All-four-flow comparison: pass. Every metric cell is baseline / peer, in microseconds.

hp/sfc-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs
nos_er	pass	17.579 / 20.810	24.674 / 33.254	31.779 / 38.274	143.285 / 167.740
md	pass	34.546 / 24.991	41.986 / 37.905	54.767 / 43.705	143.017 / 97.064
md_nos_er	pass	54.092 / 42.023	65.184 / 57.004	85.433 / 72.311	146.394 / 215.297
tick_to_trade	pass	13.824 / 18.456	20.764 / 25.615	28.484 / 29.063	113.314 / 107.140

scan/mlx5-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs
nos_er	pass	28.344 / 30.570	44.822 / 57.708	54.104 / 64.678	150.401 / 124.935
md	pass	50.956 / 36.103	68.869 / 62.419	80.620 / 68.780	158.862 / 167.502
md_nos_er	pass	82.433 / 62.158	107.769 / 97.252	121.973 / 149.359	176.256 / 187.832
tick_to_trade	pass	22.795 / 28.659	39.560 / 44.810	44.944 / 50.969	59.215 / 147.938

Complete matched report JSON → · Immutable retention seal →

Java

LibHFT Java Pure callback / Philadelphia-fast callback

All-four-flow comparison: blocked. Every metric cell is baseline / peer, in microseconds.

hp/sfc-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	blocked	—	—	—	—	baseline_report_gate; peer_report_gate
md	blocked	—	—	—	—	baseline_report_gate
md_nos_er	blocked	—	—	—	—	baseline_report_gate
tick_to_trade	blocked	—	—	—	—	baseline_report_gate; peer_report_gate

scan/mlx5-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	blocked	—	—	—	—	baseline_report_gate; peer_report_gate
md	blocked	—	—	—	—	baseline_report_gate; peer_report_gate
md_nos_er	blocked	—	—	—	—	baseline_report_gate; peer_report_gate
tick_to_trade	blocked	—	—	—	—	baseline_report_gate; peer_report_gate

Complete matched report JSON → · Immutable retention seal →

.NET

LibHFT public managed .NET 10, first CPU / QuickFIX/n parsing-only .NET 10, first CPU

All-four-flow comparison: blocked. Every metric cell is baseline / peer, in microseconds.

hp/sfc-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	pass	21.736 / 62.670	27.822 / 65.294	34.552 / 209.883	669.794 / 6217.287
md	blocked	—	—	—	—	peer:achieved_rate_below_98pct
md_nos_er	blocked	—	—	—	—	peer:achieved_rate_below_98pct;scheduler_delay_above_interval
tick_to_trade	pass	20.222 / 47.910	24.930 / 50.210	31.631 / 196.593	39.938 / 6134.838

scan/mlx5-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	blocked	—	—	—	—	peer:achieved_rate_below_98pct
md	blocked	—	—	—	—	peer:achieved_rate_below_98pct;scheduler_delay_above_interval
md_nos_er	blocked	—	—	—	—	peer:achieved_rate_below_98pct;scheduler_delay_above_interval
tick_to_trade	blocked	—	—	—	—	peer:achieved_rate_below_98pct

Complete matched report JSON → · Physical collection and independent replay → · Source 41f8706e3354, SHA-256 ec8305a6ce7f7462ef221b519803ecff6a84015e2bd52791347293104e6fe99d

Java

LibHFT public Java callback replies, split CPUs / Philadelphia-fast callback, split CPUs

All-four-flow comparison: blocked. Every metric cell is baseline / peer, in microseconds.

hp/sfc-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	blocked	—	—	—	—	baseline_report_gate
md	pass	52.187 / 24.669	116.032 / 45.076	158.225 / 73.763	11260.111 / 1986.154
md_nos_er	blocked	—	—	—	—	baseline_report_gate; peer_report_gate
tick_to_trade	pass	30.930 / 15.765	75.084 / 30.729	128.242 / 49.438	6133.081 / 132.984

scan/mlx5-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	blocked	—	—	—	—	baseline_report_gate
md	blocked	—	—	—	—	baseline_report_gate
md_nos_er	blocked	—	—	—	—	baseline_report_gate; peer_report_gate
tick_to_trade	blocked	—	—	—	—	baseline_report_gate; peer_report_gate

Complete matched report JSON → · Physical collection and independent replay → · Source 41f8706e3354, SHA-256 a91e30342276f29d75189b8779966fa877c660cd03f6d87227fd8de7607a444f

Java

LibHFT public Java callback replies, split CPUs (previous source) / Philadelphia-fast callback, split CPUs

All-four-flow comparison: blocked. Every metric cell is baseline / peer, in microseconds.

hp/sfc-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	blocked	—	—	—	—	java-callback-source-superseded-20261017
md	blocked	—	—	—	—	java-callback-source-superseded-20261017
md_nos_er	blocked	—	—	—	—	baseline_report_gate; peer_report_gate; java-callback-source-superseded-20261017
tick_to_trade	blocked	—	—	—	—	java-callback-source-superseded-20261017

scan/mlx5-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs	Blockers
nos_er	blocked	—	—	—	—	baseline_report_gate; java-callback-source-superseded-20261017
md	blocked	—	—	—	—	baseline_report_gate; java-callback-source-superseded-20261017
md_nos_er	blocked	—	—	—	—	baseline_report_gate; peer_report_gate; java-callback-source-superseded-20261017
tick_to_trade	blocked	—	—	—	—	baseline_report_gate; java-callback-source-superseded-20261017

Superseded: newer passing compact cells replace this source. The independently replayed numeric comparison has its own source, products and failed gates. Original archived grades and bytes are preserved. Supersession record →

Complete matched report JSON → · Physical collection and independent replay → · Source 41f8706e3354, SHA-256 3a6b279b15b7ab090f4f9cab84831309b432bb4c68785e3d0c9da68595e46cfd

Rust

LibHFT Rust complete RTT / callback, first CPU / NanoFIX complete RTT / callback, first CPU

All-four-flow comparison: pass. Every metric cell is baseline / peer, in microseconds.

hp/sfc-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs
nos_er	pass	21.347 / 21.466	22.896 / 23.060	28.593 / 29.529	41.543 / 56.536
md	pass	30.561 / 29.976	31.999 / 31.142	39.456 / 38.560	52.746 / 62.526
md_nos_er	pass	51.890 / 47.435	53.562 / 49.005	63.768 / 58.790	72.645 / 112.333
tick_to_trade	pass	14.874 / 19.988	16.312 / 21.353	23.107 / 28.146	30.614 / 34.026

scan/mlx5-DAC/same-host-netns

Flow	Matched gate	p50 µs	p99 µs	p99.9 µs	Worst µs
nos_er	pass	32.280 / 34.483	46.600 / 43.940	50.617 / 52.207	76.356 / 100.733
md	pass	44.513 / 45.170	59.395 / 54.750	63.832 / 63.206	85.434 / 98.521
md_nos_er	pass	76.906 / 68.818	94.441 / 91.549	103.380 / 97.068	116.330 / 143.910
tick_to_trade	pass	23.770 / 32.197	38.220 / 39.382	41.050 / 48.067	48.390 / 57.741

Complete matched report JSON → · Physical collection and independent replay → · Source 41f8706e3354, SHA-256 9af1f97d8c628afbf12ce2e6776d114fb9e15fc72f0b3dd8ce0b51be2a192df8