Performance

Performance

Throughput and footprint of Titanium as a reverse / edge proxy and as a decrypting (MITM) proxy, measured on the same harness against YARP, nginx, HAProxy, and Envoy where each OS can run them.

RPS is requests per second (gRPC charts use RPC/s). Numbers are Release builds on matched GitHub-hosted runners: Windows and Linux at 4 vCPU / 16 GiB, macOS at macos-15-intel 4-core / 14 GB. Read within one chart or table — absolute RPS is not comparable across operating systems. Not possible means that product cannot run that path on that OS; Not measured means the path exists but no published number yet.

For pooling knobs and certificate first-visit tuning, see Performance and pooling. Laptop cool A/B tables (not publishable) live on Performance Local Lab.

Why this comparison is fair

• Same load generator, same origin process, and the same warmup / measure windows (2s / 8s) with the same concurrency ramp (8, 16, 32, 64).
• Every reverse arm is three OS processes: load generator + origin + proxy. Origin-direct omits the proxy; peers are never in-process with the client.
• Same runner class per table (windows-latest / ubuntu-latest / macos-15-intel). Laptop numbers are never mixed into these tables.
• Peers use equivalent TLS/ALPN and streaming-friendly settings on the same loopback shape. HAProxy and Envoy are Linux/macOS only — Windows cells are Not possible.
• MITM (HTTPS decryption with forged certificates) is Titanium-only; peers cannot MITM. Those tables show Titanium MITM overhead versus its own reverse path on the same wires.
• Tiny keep-alive GET (~56-byte JSON) is the industry RPS shape (same class as wrk / TechEmpower). It is also real for small JSON APIs and health checks.
• Same-protocol H2↔H2 / H3↔H3 on that shape is Titanium’s best case: with interception off, Titanium copies frames instead of decoding and re-encoding headers (peers do a full HTTP decode). Medals there are not the typical reverse-proxy job.
• Typical reverse is H1 TLS→H1 or H2→H1 (~1.1× YARP on tiny GET). With larger bodies, see Heavier reverse: at 64 KiB H2 TLS→H2 TLS, Titanium is behind YARP (~0.69–0.76×).

How to read the tables

• Sustain = last concurrency that still met error/latency SLOs. Peak = highest RPS in that ramp.
• 🥇 = best among Titanium / nginx / YARP on Reverse rows (highest RPS; on a tie, lower memory then lower CPU%). Gold medals are per cell on tiny GET — an H2↔H2 gold is that frame-copy best case, not “Titanium is 1.7× on all reverse.” For larger-body H2→H2, see the heavier tables.
• MITM tables are Titanium-only. Lite÷Reverse / Full÷Reverse = Titanium MITM sustain ÷ Titanium reverse sustain on the same Client×Origin pair — the overhead of decrypting and intercepting versus bare reverse, not versus nginx or YARP.
• Not possible = cannot do that path. Not measured = path exists but no published number yet.

Contents

• Why this comparison is fair
• How to read the tables
• Measurement environment
  • Windows (GitHub-hosted windows-latest)
  • Linux (GitHub-hosted ubuntu-latest)
  • macOS (GitHub-hosted macos-15-intel)
  • Saturation control
• Windows — Titanium vs nginx vs HAProxy vs Envoy vs YARP
• Linux — Titanium vs nginx vs HAProxy vs Envoy vs YARP
• macOS — Titanium vs nginx vs HAProxy vs Envoy vs YARP
• Editions (CLI / Plus / Intercept)
• Cross-version (7.0 vs 6.0)
• Heavier reverse workloads
• Unary gRPC (H2 TLS)
• Other measurements
• Raising limits on large hosts
• Maintainer notes

Measurement environment

All three OS use the 4-core-class public-repo GitHub-hosted runners: Windows / Linux at 4 vCPU / 16 GiB / 14 GB SSD, macOS at macos-15-intel 4-core / 14 GB (not macos-latest). Same harness knobs: warmup 2s / measure 8s; concurrency 8, 16, 32, 64; median of 3 repeats. Prefer Titanium÷YARP / Titanium÷nginx ratios over absolute RPS.

Laptop High-perf / cool-paired Windows numbers live on Performance Local Lab. Do not mix those absolutes into the tables below.

Windows (GitHub-hosted windows-latest)

OS	Windows Server (GitHub-hosted windows-latest)
CPU	4 logical processors
RAM	16 GiB
Runtime	.NET 10.0.x
nginx	nginx/Windows 1.31.3 (same-OS only; no QUIC)
HAProxy	Not possible on Windows (no official port)
Envoy	Not possible on Windows (upstream discontinued Windows builds)
YARP	Yarp.ReverseProxy 2.3.0
Harness	RpsLoadProbe Release; median of 3 repeats

Linux (GitHub-hosted ubuntu-latest)

OS	Ubuntu 24.04.x LTS
CPU	4 logical processors (AMD EPYC; runners this pass were 7763 / 9V74)
RAM	16 GiB
Runtime	.NET 10.0.11
nginx	nginx/1.31.4 (nginx.org mainline, --with-http_v3_module)
HAProxy	HAProxy 3.2.23 built with USE_QUIC (GHA; Ubuntu distro 2.8 is not QUIC-capable)
Envoy	pinned GitHub release static binary 1.36.7 (HTTP/3 compiled in)
YARP	Yarp.ReverseProxy 2.3.0
Harness	RpsLoadProbe Release; median of 3 repeats where noted

macOS (GitHub-hosted macos-15-intel)

OS	macOS 15 (GitHub-hosted macos-15-intel, Intel x86_64)
CPU	4 logical processors
RAM	14 GB
Runtime	.NET 10.0.x
nginx	Homebrew nginx with --with-http_v3_module (workflow fails if missing)
HAProxy	Homebrew haproxy with USE_QUIC (workflow fails if missing; 3.2.23 osx source fallback)
Envoy	Homebrew bottle when present; else pinned darwin-amd64 1.36.7 (official GitHub assets are Linux-only). HTTP/3 compiled in.
MsQuic	Homebrew libmsquic + openssl@3 on DYLD_LIBRARY_PATH / DYLD_FALLBACK_LIBRARY_PATH (QuicListener.IsSupported)
YARP	Yarp.ReverseProxy 2.3.0
Harness	RpsLoadProbe Release; median of 3 repeats where noted

Do not use macos-latest (Apple Silicon, 3-core / 7 GB) for publishable saturation numbers.

Saturation control

Calibration for the shared 4 vCPU loopback shape: how close client + origin are to saturated before ranking reverse peers. Tiny keep-alive GET. Median of 3 repeats @ 9a2b3a1e — 34441539402. Warmup 2s / measure 8s; concurrency 8, 16, 32, 64. Block A % of origin-HttpClient uses median peak RPS. Blocks B/C use peer÷YARP / ÷nginx on median peak (not % of H1 origin). RPS cells embed median RSS / CPU for the proxy child plus its full descendant tree (serve-proxy → nginx master → workers); origin-direct samples the origin child. Product matrices below use matched dotnet-httpclient only (not bombardier).

Block A — H1 plain

Arm	Generator	Notes
origin-direct	dotnet-httpclient	No proxy — origin ceiling under the same client used for publishable ratios
origin-direct-bombardier	bombardier	External client check (CI installs bombardier)
bare-reverse-http1	dotnet-httpclient	Thin C# H1 reverse (BareHttp1ReverseProxy) — .NET runtime / loopback ceiling for a three-process reverse hop; not a product peer
nginx-reverse-http1 / yarp-reverse-http1 / twp-reverse-http1	dotnet-httpclient	Product peers (medals among these three only)

Windows (windows-latest)

Arm	Generator	Sustain	Peak	% of origin-HttpClient
origin-direct	dotnet-httpclient	50,319 (55 MiB / 40.4% CPU)	50,319 (55 MiB / 40.4% CPU)	100.0%
origin-direct-bombardier	bombardier	38,565 (56 MiB / 21.8% CPU)	38,565 (56 MiB / 21.8% CPU)	76.6%
bare-reverse-http1	dotnet-httpclient	25,016 (58 MiB / 43.8% CPU)	25,016 (58 MiB / 43.8% CPU)	49.7%
nginx-reverse-http1	dotnet-httpclient	13,405 (125 MiB / 24.8% CPU)	13,405 (125 MiB / 24.8% CPU)	26.6%
yarp-reverse-http1	dotnet-httpclient	20,802 (87 MiB / 49.0% CPU)	20,802 (87 MiB / 49.0% CPU)	41.3%
twp-reverse-http1	dotnet-httpclient	🥇 24,558 (75 MiB / 48.0% CPU)	24,558 (75 MiB / 48.0% CPU)	48.8%

Linux (ubuntu-latest)

Arm	Generator	Sustain	Peak	% of origin-HttpClient
origin-direct	dotnet-httpclient	102,038 (80 MiB / 44.2% CPU)	102,038 (80 MiB / 44.2% CPU)	100.0%
origin-direct-bombardier	bombardier	61,434 (80 MiB / 37.4% CPU)	61,434 (80 MiB / 37.4% CPU)	60.2%
bare-reverse-http1	dotnet-httpclient	46,442 (70 MiB / 46.0% CPU)	46,442 (70 MiB / 46.0% CPU)	45.5%
nginx-reverse-http1	dotnet-httpclient	🥇 56,585 (76 MiB / 40.3% CPU)	56,585 (76 MiB / 40.3% CPU)	55.5%
yarp-reverse-http1	dotnet-httpclient	41,835 (116 MiB / 49.3% CPU)	41,835 (116 MiB / 49.3% CPU)	41.0%
twp-reverse-http1	dotnet-httpclient	47,721 (95 MiB / 50.9% CPU)	47,721 (95 MiB / 50.9% CPU)	46.8%

Reverse peers are about 50–46% of the origin-direct HttpClient peak on this runner class (Win TWP 50.0%, Lin TWP 45.6%). Prefer the % column over absolute RPS across runs. Bare and origin-direct are controls (not medal peers).

Block B — H2 TLS→H1

Peer ratios (÷YARP / ÷nginx) on median peak; RPS cells embed (MiB / CPU%).

Windows (windows-latest)

Arm	Generator	Sustain	Peak	÷YARP	÷nginx
nginx-reverse-http2	dotnet-httpclient	7,992 (141 MiB / 24.6% CPU)	7,992 (141 MiB / 24.6% CPU)	0.28×	1.00×
yarp-reverse-http2	dotnet-httpclient	28,294 (97 MiB / 54.9% CPU)	28,294 (97 MiB / 54.9% CPU)	1.00×	3.54×
twp-reverse-http2-cleartext	dotnet-httpclient	🥇 33,720 (105 MiB / 52.2% CPU)	33,720 (105 MiB / 52.2% CPU)	1.19×	4.22×

Linux (ubuntu-latest)

Arm	Generator	Sustain	Peak	÷YARP	÷nginx
nginx-reverse-http2	dotnet-httpclient	23,276 (102 MiB / 18.9% CPU)	23,276 (102 MiB / 18.9% CPU)	0.53×	1.00×
yarp-reverse-http2	dotnet-httpclient	44,299 (122 MiB / 48.0% CPU)	44,299 (122 MiB / 48.0% CPU)	1.00×	1.90×
twp-reverse-http2-cleartext	dotnet-httpclient	🥇 49,167 (124 MiB / 52.1% CPU)	49,167 (124 MiB / 52.1% CPU)	1.11×	2.11×

Block C — H3→H1

Same layout as Block B. Requires QuicListener. nginx needs http_v3_module (Windows nginx has no QUIC). HAProxy needs USE_QUIC (GHA Linux/macOS require it). Envoy 1.20+ includes HTTP/3.

Windows (windows-latest)

Arm	Generator	Sustain	Peak	÷YARP	÷nginx
nginx-reverse-http3-cleartext	dotnet-httpclient	Not possible (no QUIC)	Not possible (no QUIC)	—	—
yarp-reverse-http3-cleartext	dotnet-httpclient	14,041 (142 MiB / 51.8% CPU)	14,041 (142 MiB / 51.8% CPU)	1.00×	—
twp-reverse-http3-cleartext	dotnet-httpclient	🥇 14,348 (104 MiB / 43.8% CPU)	14,348 (104 MiB / 43.8% CPU)	1.02×	—

Linux (ubuntu-latest)

Arm	Generator	Sustain	Peak	÷YARP	÷nginx
nginx-reverse-http3-cleartext	dotnet-httpclient	0 (110 MiB / 21.8% CPU)	24,928 (110 MiB / 21.8% CPU)	0.89×	1.00×
yarp-reverse-http3-cleartext	dotnet-httpclient	27,936 (195 MiB / 48.8% CPU)	27,936 (195 MiB / 48.8% CPU)	1.00×	1.12×
twp-reverse-http3-cleartext	dotnet-httpclient	🥇 30,636 (159 MiB / 52.7% CPU)	30,636 (159 MiB / 52.7% CPU)	1.10×	1.23×

Windows — Titanium vs nginx vs HAProxy vs Envoy vs YARP

Client / origin: HTTP version and whether TLS is used (plain = cleartext, TLS = encrypted, QUIC = HTTP/3).

Reverse

Median of 3 repeats on windows-latest (4 vCPU / 16 GiB). Bare reverse 5×5 @ 9a2b3a1e — compare-product 34441526151. Warmup 2s / measure 8s; concurrency 8, 16, 32, 64. Prefer TWP÷peer ratios over absolute RPS. RPS cells include median RSS / CPU at the peak-RPS step as <br><sub>(MiB / CPU%)</sub>. nginx terminate peers use keepalive 256 + streaming buffers. HAProxy / Envoy are Linux-only peers (no official Windows port). Laptop High-perf / cool-paired numbers stay on the local lab. Product 5×5 is ~56-byte JSON keep-alive GET; H2/H3 same-protocol cells are mostly header work with a tiny body (Titanium best case) — see Why this comparison is fair.

Load generators: Reverse inbound H3 arms use dotnet-httpclient (http_version=3.0, RequestVersionExact). nginx/Windows is same-OS only (no QUIC). HAProxy/Envoy are Linux-only terminate peers.

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
HTTP/1 · plain	HTTP/1 · plain	🥇 23134 (75 MiB / 47.5% CPU)	🥇 23134 (75 MiB / 47.5% CPU)	13916 (125 MiB / 24.9% CPU)	13916 (125 MiB / 24.9% CPU)	Not possible	Not possible	Not possible	Not possible	21699 (86 MiB / 48.8% CPU)	21699 (86 MiB / 48.8% CPU)
HTTP/1 · plain	HTTP/1 · TLS	🥇 20749 (89 MiB / 52.8% CPU)	🥇 20749 (89 MiB / 52.8% CPU)	8366 (135 MiB / 24.6% CPU)	8366 (135 MiB / 24.6% CPU)	Not possible	Not possible	Not possible	Not possible	18784 (100 MiB / 49% CPU)	18784 (100 MiB / 49% CPU)
HTTP/1 · plain	HTTP/2 · plain	🥇 45733 (115 MiB / 50.5% CPU)	🥇 45733 (115 MiB / 50.5% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	Not possible	Not possible	Not possible	Not possible	39820 (90 MiB / 49.2% CPU)	39820 (90 MiB / 49.2% CPU)
HTTP/1 · plain	HTTP/2 · TLS	🥇 32162 (120 MiB / 46.7% CPU)	🥇 32162 (120 MiB / 46.7% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	Not possible	Not possible	Not possible	Not possible	29670 (98 MiB / 49.7% CPU)	29670 (98 MiB / 49.7% CPU)
HTTP/1 · plain	HTTP/3 · QUIC	🥇 18315 (106 MiB / 50.6% CPU)	🥇 18315 (106 MiB / 50.6% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not possible	Not possible	Not possible	Not possible	17823 (117 MiB / 51% CPU)	17823 (117 MiB / 51% CPU)
HTTP/1 · TLS	HTTP/1 · plain	🥇 25720 (90 MiB / 48.4% CPU)	🥇 25720 (90 MiB / 48.4% CPU)	12987 (141 MiB / 24.8% CPU)	12987 (141 MiB / 24.8% CPU)	Not possible	Not possible	Not possible	Not possible	22690 (102 MiB / 48.2% CPU)	22690 (102 MiB / 48.2% CPU)
HTTP/1 · TLS	HTTP/1 · TLS	🥇 18942 (91 MiB / 48.6% CPU)	🥇 18942 (91 MiB / 48.6% CPU)	7218 (143 MiB / 24.8% CPU)	7218 (143 MiB / 24.8% CPU)	Not possible	Not possible	Not possible	Not possible	17296 (104 MiB / 49.7% CPU)	17296 (104 MiB / 49.7% CPU)
HTTP/1 · TLS	HTTP/2 · plain	🥇 27614 (111 MiB / 44.5% CPU)	🥇 27614 (111 MiB / 44.5% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	Not possible	Not possible	Not possible	Not possible	26015 (104 MiB / 47.6% CPU)	26015 (104 MiB / 47.6% CPU)
HTTP/1 · TLS	HTTP/2 · TLS	🥇 32132 (117 MiB / 45.4% CPU)	🥇 32132 (117 MiB / 45.4% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	Not possible	Not possible	Not possible	Not possible	29999 (103 MiB / 48.2% CPU)	29999 (103 MiB / 48.2% CPU)
HTTP/1 · TLS	HTTP/3 · QUIC	🥇 16275 (110 MiB / 51.3% CPU)	🥇 16275 (110 MiB / 51.3% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not possible	Not possible	Not possible	Not possible	15797 (124 MiB / 51.4% CPU)	15797 (124 MiB / 51.4% CPU)
HTTP/2 · plain	HTTP/1 · plain	🥇 34933 (89 MiB / 55% CPU)	🥇 34933 (89 MiB / 55% CPU)	9362 (127 MiB / 24.4% CPU)	9362 (127 MiB / 24.4% CPU)	Not possible	Not possible	Not possible	Not possible	31625 (86 MiB / 54.4% CPU)	31625 (86 MiB / 54.4% CPU)
HTTP/2 · plain	HTTP/1 · TLS	🥇 35086 (105 MiB / 49% CPU)	🥇 35086 (105 MiB / 49% CPU)	8849 (138 MiB / 24.9% CPU)	8849 (138 MiB / 24.9% CPU)	Not possible	Not possible	Not possible	Not possible	32740 (92 MiB / 50.1% CPU)	32740 (92 MiB / 50.1% CPU)
HTTP/2 · plain	HTTP/2 · plain	🥇 112638 (61 MiB / 28.3% CPU)	🥇 112638 (61 MiB / 28.3% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	Not possible	Not possible	Not possible	Not possible	64290 (90 MiB / 49.6% CPU)	64290 (90 MiB / 49.6% CPU)
HTTP/2 · plain	HTTP/2 · TLS	🥇 70105 (62 MiB / 20.1% CPU)	🥇 70105 (62 MiB / 20.1% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	Not possible	Not possible	Not possible	Not possible	43971 (99 MiB / 35.9% CPU)	43971 (99 MiB / 35.9% CPU)
HTTP/2 · plain	HTTP/3 · QUIC	🥇 33088 (128 MiB / 53.1% CPU)	🥇 33088 (128 MiB / 53.1% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not possible	Not possible	Not possible	Not possible	31406 (130 MiB / 52% CPU)	31406 (130 MiB / 52% CPU)
HTTP/2 · TLS	HTTP/1 · plain	🥇 34943 (97 MiB / 52.2% CPU)	🥇 34943 (97 MiB / 52.2% CPU)	8430 (141 MiB / 24.8% CPU)	8430 (141 MiB / 24.8% CPU)	Not possible	Not possible	Not possible	Not possible	29441 (96 MiB / 53.8% CPU)	29441 (96 MiB / 53.8% CPU)
HTTP/2 · TLS	HTTP/1 · TLS	🥇 29487 (97 MiB / 53.6% CPU)	🥇 29487 (97 MiB / 53.6% CPU)	6542 (144 MiB / 24.6% CPU)	6542 (144 MiB / 24.6% CPU)	Not possible	Not possible	Not possible	Not possible	25117 (98 MiB / 54.2% CPU)	25117 (98 MiB / 54.2% CPU)
HTTP/2 · TLS	HTTP/2 · plain	🥇 107152 (75 MiB / 29.1% CPU)	🥇 107152 (75 MiB / 29.1% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	Not possible	Not possible	Not possible	Not possible	59098 (102 MiB / 52.7% CPU)	59098 (102 MiB / 52.7% CPU)
HTTP/2 · TLS	HTTP/2 · TLS	🥇 84487 (73 MiB / 28.8% CPU)	🥇 84487 (73 MiB / 28.8% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	Not possible	Not possible	Not possible	Not possible	49546 (98 MiB / 49% CPU)	49546 (98 MiB / 49% CPU)
HTTP/2 · TLS	HTTP/3 · QUIC	🥇 30034 (127 MiB / 53.4% CPU)	🥇 30034 (127 MiB / 53.4% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not possible	Not possible	Not possible	Not possible	25937 (123 MiB / 51.5% CPU)	25937 (123 MiB / 51.5% CPU)
HTTP/3 · QUIC	HTTP/1 · plain	🥇 19348 (107 MiB / 44% CPU)	🥇 19348 (107 MiB / 44% CPU)	Not measured	Not measured	Not possible	Not possible	Not possible	Not possible	18041 (143 MiB / 50.9% CPU)	18041 (143 MiB / 50.9% CPU)
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 13646 (113 MiB / 46.7% CPU)	🥇 13646 (113 MiB / 46.7% CPU)	Not measured	Not measured	Not possible	Not possible	Not possible	Not possible	13437 (145 MiB / 52.4% CPU)	13437 (145 MiB / 52.4% CPU)
HTTP/3 · QUIC	HTTP/2 · plain	🥇 34128 (136 MiB / 47.6% CPU)	🥇 34128 (136 MiB / 47.6% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	Not possible	Not possible	Not possible	Not possible	25450 (166 MiB / 50.1% CPU)	25450 (166 MiB / 50.1% CPU)
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 33850 (138 MiB / 48% CPU)	🥇 33850 (138 MiB / 48% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	Not possible	Not possible	Not possible	Not possible	26828 (168 MiB / 48.5% CPU)	26828 (168 MiB / 48.5% CPU)
HTTP/3 · QUIC	HTTP/3 · QUIC	🥇 23150 (121 MiB / 41.8% CPU)	🥇 23150 (121 MiB / 41.8% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not possible	Not possible	Not possible	Not possible	12064 (167 MiB / 53.1% CPU)	12064 (167 MiB / 53.1% CPU)

MITM (TWP only)

Same Client×Origin wires with interception on (compare-product 34441526151). Lite = no-op handlers (unchanged-lite finish). Full = append-only header mutation (harness: one probe header each way; product: generic append-only relay via MitmCompressedRelayHelper). nginx/HAProxy/Envoy/YARP cannot MITM. Lite÷Reverse / Full÷Reverse vs bare reverse (same job / comparison-group shard). Completion gate: Lite ≥ 0.50× and Full ≥ 0.50× reverse sustain @ c=64 (median of 3 GHA runs); reverse TWP÷YARP ≥ 0.70× (no terminate-peer gate).

v1 append-only relay (2026-08-27): Pre-fix H2→H2 Full÷Reverse was 0.13–0.16× (32960766249). Post-fix @ df172718: H2 plain→H2 plain Full 0.77–0.79×, H3→H1 Full 0.91–0.93×, all MITM arms ≥ 0.70× on median of 33041445371, 33055267086, 33055272140.

v2 drop-only + non-unique append (2026-08-27): MitmStaticRebuildHelper rebuilds static HPACK/QPACK after 1–4 unique header drops; trailing non-unique appends stay on compressed relay. @ 9a2b3a1e: all MITM arms ≥ 0.70× on GHA median (33087088466, 33087091622, 33105885748 Linux; 33087085235, 33087088466, 33087091622 Windows). H2 plain→H2 plain Full 0.77–0.79× (Win) / 0.78× (Lin).

Client	Origin	Lite sustain	Full sustain	Lite÷Reverse	Full÷Reverse
HTTP/1 · plain	HTTP/1 · plain	25054 (82 MiB / 50.4% CPU)	24740 (82 MiB / 51.6% CPU)	1.08×	1.07×
HTTP/1 · plain	HTTP/1 · TLS	20351 (95 MiB / 54.2% CPU)	19992 (95 MiB / 52% CPU)	0.98×	0.96×
HTTP/1 · plain	HTTP/2 · plain	45468 (110 MiB / 51.5% CPU)	43600 (120 MiB / 49.4% CPU)	0.99×	0.95×
HTTP/1 · plain	HTTP/2 · TLS	32619 (118 MiB / 46.8% CPU)	31475 (125 MiB / 46.3% CPU)	1.01×	0.98×
HTTP/1 · plain	HTTP/3 · QUIC	17858 (105 MiB / 47.7% CPU)	17488 (104 MiB / 52.6% CPU)	0.98×	0.95×
HTTP/1 · TLS	HTTP/1 · plain	24388 (98 MiB / 47.9% CPU)	24773 (95 MiB / 48.2% CPU)	0.95×	0.96×
HTTP/1 · TLS	HTTP/1 · TLS	18766 (94 MiB / 49.4% CPU)	18334 (96 MiB / 49.4% CPU)	0.99×	0.97×
HTTP/1 · TLS	HTTP/2 · plain	27503 (112 MiB / 47.3% CPU)	27245 (112 MiB / 46.1% CPU)	1×	0.99×
HTTP/1 · TLS	HTTP/2 · TLS	31677 (118 MiB / 47.9% CPU)	30700 (117 MiB / 44.2% CPU)	0.99×	0.96×
HTTP/1 · TLS	HTTP/3 · QUIC	15862 (113 MiB / 50.9% CPU)	15726 (112 MiB / 50.8% CPU)	0.97×	0.97×
HTTP/2 · plain	HTTP/1 · plain	34123 (95 MiB / 55.4% CPU)	33304 (96 MiB / 56.8% CPU)	0.98×	0.95×
HTTP/2 · plain	HTTP/1 · TLS	34643 (106 MiB / 50.5% CPU)	34092 (109 MiB / 52.9% CPU)	0.99×	0.97×
HTTP/2 · plain	HTTP/2 · plain	86231 (71 MiB / 41.3% CPU)	80246 (73 MiB / 41.9% CPU)	0.77×	0.71×
HTTP/2 · plain	HTTP/2 · TLS	72044 (77 MiB / 36.5% CPU)	67973 (77 MiB / 39.7% CPU)	1.03×	0.97×
HTTP/2 · plain	HTTP/3 · QUIC	33179 (124 MiB / 53.4% CPU)	32949 (129 MiB / 53.1% CPU)	1×	1×
HTTP/2 · TLS	HTTP/1 · plain	33820 (98 MiB / 55.2% CPU)	32652 (107 MiB / 54.9% CPU)	0.97×	0.93×
HTTP/2 · TLS	HTTP/1 · TLS	28427 (100 MiB / 56.2% CPU)	27770 (105 MiB / 53.2% CPU)	0.96×	0.94×
HTTP/2 · TLS	HTTP/2 · plain	84248 (86 MiB / 40.5% CPU)	80554 (84 MiB / 39.7% CPU)	0.79×	0.75×
HTTP/2 · TLS	HTTP/2 · TLS	68743 (86 MiB / 38.1% CPU)	65184 (81 MiB / 39.6% CPU)	0.81×	0.77×
HTTP/2 · TLS	HTTP/3 · QUIC	29843 (126 MiB / 54.4% CPU)	29276 (133 MiB / 53.5% CPU)	0.99×	0.97×
HTTP/3 · QUIC	HTTP/1 · plain	18648 (117 MiB / 44.9% CPU)	17458 (114 MiB / 44.7% CPU)	0.96×	0.9×
HTTP/3 · QUIC	HTTP/1 · TLS	12964 (123 MiB / 46.8% CPU)	12392 (116 MiB / 47.5% CPU)	0.95×	0.91×
HTTP/3 · QUIC	HTTP/2 · plain	30904 (132 MiB / 48.6% CPU)	29726 (133 MiB / 50.6% CPU)	0.91×	0.87×
HTTP/3 · QUIC	HTTP/2 · TLS	31489 (143 MiB / 48.5% CPU)	30510 (143 MiB / 48.3% CPU)	0.93×	0.9×
HTTP/3 · QUIC	HTTP/3 · QUIC	16061 (120 MiB / 49.2% CPU)	15312 (119 MiB / 49.5% CPU)	0.69×	0.66×

Linux — Titanium vs nginx vs HAProxy vs Envoy vs YARP

Reverse

Median of 3 repeats on ubuntu-latest (4 vCPU / 16 GiB). Bare reverse 5×5 @ 9a2b3a1e — compare-product 34441526151. Warmup 2s / measure 8s; concurrency 8, 16, 32, 64. Linux nginx is the authoritative nginx baseline. HAProxy (3.2 USE_QUIC) and Envoy (GitHub release, HTTP/3 compiled in) run on the same loopback shape as nginx/YARP. nginx terminate peers use keepalive 256 + streaming buffers. The RPS workflow installs nginx.org mainline (http_v3_module), a QUIC-enabled HAProxy, Envoy, and libmsquic. Prefer ratios over absolute RPS. Product 5×5 is ~56-byte JSON keep-alive GET; H2/H3 same-protocol cells are mostly header work with a tiny body (Titanium best case) — see Why this comparison is fair.

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
HTTP/1 · plain	HTTP/1 · plain	36672 (94 MiB / 50.5% CPU)	36672 (94 MiB / 50.5% CPU)	🥇 44112 (76 MiB / 40.4% CPU)	🥇 44112 (76 MiB / 40.4% CPU)	41748 (67 MiB / 41.3% CPU)	41748 (67 MiB / 41.3% CPU)	24415 (116 MiB / 59.5% CPU)	24415 (116 MiB / 59.5% CPU)	32393 (115 MiB / 48.8% CPU)	32393 (115 MiB / 48.8% CPU)
HTTP/1 · plain	HTTP/1 · TLS	28506 (107 MiB / 49.3% CPU)	28506 (107 MiB / 49.3% CPU)	🥇 34687 (93 MiB / 40.7% CPU)	🥇 34687 (93 MiB / 40.7% CPU)	32839 (71 MiB / 43% CPU)	32839 (71 MiB / 43% CPU)	22383 (117 MiB / 56.4% CPU)	22383 (117 MiB / 56.4% CPU)	25710 (138 MiB / 49.6% CPU)	25710 (138 MiB / 49.6% CPU)
HTTP/1 · plain	HTTP/2 · plain	🥇 45673 (129 MiB / 52.7% CPU)	🥇 45673 (129 MiB / 52.7% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	26830 (68 MiB / 42.8% CPU)	26830 (68 MiB / 42.8% CPU)	27447 (116 MiB / 61% CPU)	27447 (116 MiB / 61% CPU)	40965 (125 MiB / 49.4% CPU)	40965 (125 MiB / 49.4% CPU)
HTTP/1 · plain	HTTP/2 · TLS	🥇 37934 (147 MiB / 49.1% CPU)	🥇 37934 (147 MiB / 49.1% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	35699 (68 MiB / 43% CPU)	35699 (68 MiB / 43% CPU)	19949 (117 MiB / 61.9% CPU)	19949 (117 MiB / 61.9% CPU)	35520 (132 MiB / 47.3% CPU)	35520 (132 MiB / 47.3% CPU)
HTTP/1 · plain	HTTP/3 · QUIC	🥇 26151 (141 MiB / 53.8% CPU)	🥇 26151 (141 MiB / 53.8% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not measured	Not measured	12126 (120 MiB / 56.3% CPU)	12126 (120 MiB / 56.3% CPU)	24086 (155 MiB / 48.3% CPU)	24086 (155 MiB / 48.3% CPU)
HTTP/1 · TLS	HTTP/1 · plain	27742 (111 MiB / 48.9% CPU)	27742 (111 MiB / 48.9% CPU)	33232 (100 MiB / 41.5% CPU)	33232 (100 MiB / 41.5% CPU)	🥇 33283 (85 MiB / 43.6% CPU)	🥇 33283 (85 MiB / 43.6% CPU)	21660 (127 MiB / 56.4% CPU)	21660 (127 MiB / 56.4% CPU)	24735 (142 MiB / 49.4% CPU)	24735 (142 MiB / 49.4% CPU)
HTTP/1 · TLS	HTTP/1 · TLS	23878 (111 MiB / 48.2% CPU)	23878 (111 MiB / 48.2% CPU)	🥇 28042 (103 MiB / 41.6% CPU)	🥇 28042 (103 MiB / 41.6% CPU)	27394 (86 MiB / 43.8% CPU)	27394 (86 MiB / 43.8% CPU)	19309 (128 MiB / 55% CPU)	19309 (128 MiB / 55% CPU)	21716 (142 MiB / 49.4% CPU)	21716 (142 MiB / 49.4% CPU)
HTTP/1 · TLS	HTTP/2 · plain	🥇 34010 (140 MiB / 51.4% CPU)	🥇 34010 (140 MiB / 51.4% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	21760 (83 MiB / 42.9% CPU)	21760 (83 MiB / 42.9% CPU)	18208 (127 MiB / 58.6% CPU)	18208 (127 MiB / 58.6% CPU)	30979 (141 MiB / 49.2% CPU)	30979 (141 MiB / 49.2% CPU)
HTTP/1 · TLS	HTTP/2 · TLS	🥇 29656 (157 MiB / 48.6% CPU)	🥇 29656 (157 MiB / 48.6% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	29078 (83 MiB / 44% CPU)	29078 (83 MiB / 44% CPU)	21903 (126 MiB / 56.9% CPU)	21903 (126 MiB / 56.9% CPU)	27210 (146 MiB / 48.3% CPU)	27210 (146 MiB / 48.3% CPU)
HTTP/1 · TLS	HTTP/3 · QUIC	🥇 22269 (152 MiB / 52.9% CPU)	🥇 22269 (152 MiB / 52.9% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not measured	Not measured	11093 (130 MiB / 55.9% CPU)	11093 (130 MiB / 55.9% CPU)	19851 (166 MiB / 49.4% CPU)	19851 (166 MiB / 49.4% CPU)
HTTP/2 · plain	HTTP/1 · plain	🥇 41263 (126 MiB / 53.1% CPU)	🥇 41263 (126 MiB / 53.1% CPU)	18170 (79 MiB / 19.6% CPU)	18170 (79 MiB / 19.6% CPU)	27652 (69 MiB / 24.4% CPU)	27652 (69 MiB / 24.4% CPU)	17979 (118 MiB / 23% CPU)	17979 (118 MiB / 23% CPU)	39219 (115 MiB / 48.3% CPU)	39219 (115 MiB / 48.3% CPU)
HTTP/2 · plain	HTTP/1 · TLS	🥇 32750 (131 MiB / 51% CPU)	🥇 32750 (131 MiB / 51% CPU)	15078 (100 MiB / 19.2% CPU)	15078 (100 MiB / 19.2% CPU)	22285 (71 MiB / 24.5% CPU)	22285 (71 MiB / 24.5% CPU)	16529 (119 MiB / 23.1% CPU)	16529 (119 MiB / 23.1% CPU)	31159 (125 MiB / 48.4% CPU)	31159 (125 MiB / 48.4% CPU)
HTTP/2 · plain	HTTP/2 · plain	🥇 97239 (90 MiB / 38% CPU)	🥇 97239 (90 MiB / 38% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	30261 (69 MiB / 24.3% CPU)	30261 (69 MiB / 24.3% CPU)	24868 (116 MiB / 21.8% CPU)	24868 (116 MiB / 21.8% CPU)	55752 (122 MiB / 47.1% CPU)	55752 (122 MiB / 47.1% CPU)
HTTP/2 · plain	HTTP/2 · TLS	🥇 67702 (93 MiB / 38.2% CPU)	🥇 67702 (93 MiB / 38.2% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	23815 (68 MiB / 24.4% CPU)	23815 (68 MiB / 24.4% CPU)	16231 (118 MiB / 21.5% CPU)	16231 (118 MiB / 21.5% CPU)	44692 (129 MiB / 45.3% CPU)	44692 (129 MiB / 45.3% CPU)
HTTP/2 · plain	HTTP/3 · QUIC	🥇 30188 (150 MiB / 51.4% CPU)	🥇 30188 (150 MiB / 51.4% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not measured	Not measured	5340 (121 MiB / 24.8% CPU)	5340 (121 MiB / 24.8% CPU)	28445 (156 MiB / 45.5% CPU)	28445 (156 MiB / 45.5% CPU)
HTTP/2 · TLS	HTTP/1 · plain	🥇 38662 (128 MiB / 52.3% CPU)	🥇 38662 (128 MiB / 52.3% CPU)	17745 (100 MiB / 18.8% CPU)	17745 (100 MiB / 18.8% CPU)	24393 (81 MiB / 24.3% CPU)	24393 (81 MiB / 24.3% CPU)	17919 (128 MiB / 22.5% CPU)	17919 (128 MiB / 22.5% CPU)	34084 (122 MiB / 48.1% CPU)	34084 (122 MiB / 48.1% CPU)
HTTP/2 · TLS	HTTP/1 · TLS	🥇 30318 (119 MiB / 49.8% CPU)	🥇 30318 (119 MiB / 49.8% CPU)	15083 (110 MiB / 19.6% CPU)	15083 (110 MiB / 19.6% CPU)	20646 (85 MiB / 24.4% CPU)	20646 (85 MiB / 24.4% CPU)	16394 (128 MiB / 22.2% CPU)	16394 (128 MiB / 22.2% CPU)	27788 (131 MiB / 48.5% CPU)	27788 (131 MiB / 48.5% CPU)
HTTP/2 · TLS	HTTP/2 · plain	🥇 85544 (102 MiB / 38% CPU)	🥇 85544 (102 MiB / 38% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	53648 (85 MiB / 24.2% CPU)	53648 (85 MiB / 24.2% CPU)	24277 (126 MiB / 21.5% CPU)	24277 (126 MiB / 21.5% CPU)	44751 (125 MiB / 46.4% CPU)	44751 (125 MiB / 46.4% CPU)
HTTP/2 · TLS	HTTP/2 · TLS	🥇 63379 (99 MiB / 36% CPU)	🥇 63379 (99 MiB / 36% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	22388 (84 MiB / 24.4% CPU)	22388 (84 MiB / 24.4% CPU)	21510 (125 MiB / 20.6% CPU)	21510 (125 MiB / 20.6% CPU)	38643 (128 MiB / 45.3% CPU)	38643 (128 MiB / 45.3% CPU)
HTTP/2 · TLS	HTTP/3 · QUIC	🥇 28423 (151 MiB / 50.5% CPU)	🥇 28423 (151 MiB / 50.5% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not measured	Not measured	5476 (129 MiB / 24.6% CPU)	5476 (129 MiB / 24.6% CPU)	25085 (163 MiB / 45.8% CPU)	25085 (163 MiB / 45.8% CPU)
HTTP/3 · QUIC	HTTP/1 · plain	23715 (149 MiB / 52.2% CPU)	23715 (149 MiB / 52.2% CPU)	0 (107 MiB / 21.6% CPU)	19206 (107 MiB / 21.6% CPU)	🥇 30835 (86 MiB / 25.2% CPU)	🥇 30835 (86 MiB / 25.2% CPU)	3 (130 MiB / 0.1% CPU)	3 (130 MiB / 0.1% CPU)	21546 (186 MiB / 49% CPU)	21546 (186 MiB / 49% CPU)
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 19505 (160 MiB / 49.3% CPU)	🥇 19505 (160 MiB / 49.3% CPU)	0 (117 MiB / 22.7% CPU)	14850 (117 MiB / 22.7% CPU)	Not measured	Not measured	Not measured	Not measured	18423 (197 MiB / 49.9% CPU)	18423 (197 MiB / 49.9% CPU)
HTTP/3 · QUIC	HTTP/2 · plain	🥇 31386 (160 MiB / 56.2% CPU)	🥇 31386 (160 MiB / 56.2% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	29864 (86 MiB / 25.4% CPU)	29864 (86 MiB / 25.4% CPU)	Not measured	Not measured	26924 (197 MiB / 48% CPU)	26924 (197 MiB / 48% CPU)
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 27669 (154 MiB / 52.3% CPU)	🥇 27669 (154 MiB / 52.3% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	Not measured	Not measured	Not measured	Not measured	24488 (199 MiB / 47.6% CPU)	24488 (199 MiB / 47.6% CPU)
HTTP/3 · QUIC	HTTP/3 · QUIC	🥇 21858 (164 MiB / 47.9% CPU)	🥇 21858 (164 MiB / 47.9% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not measured	Not measured	Not measured	Not measured	17840 (209 MiB / 47% CPU)	17840 (209 MiB / 47% CPU)

MITM (TWP only)

Same Client×Origin wires with interception on (compare-product 34441526151). Lite = no-op handlers (unchanged-lite finish). Full = append-only header mutation (harness: one probe header each way; product: generic append-only relay via MitmCompressedRelayHelper). nginx/HAProxy/Envoy/YARP cannot MITM. Lite÷Reverse / Full÷Reverse vs bare reverse (same job / comparison-group shard). Completion gate: Lite ≥ 0.50× and Full ≥ 0.50× reverse sustain @ c=64 (median of 3 GHA runs); reverse TWP÷YARP ≥ 0.70× (no terminate-peer gate).

v1 append-only relay (2026-08-27): Pre-fix H2→H2 Full÷Reverse was 0.13–0.16× (32960766249). Post-fix @ df172718: H2 plain→H2 plain Full 0.77–0.79×, H3→H1 Full 0.91–0.93×, all MITM arms ≥ 0.70× on median of 33041445371, 33055267086, 33055272140.

v2 drop-only + non-unique append (2026-08-27): MitmStaticRebuildHelper rebuilds static HPACK/QPACK after 1–4 unique header drops; trailing non-unique appends stay on compressed relay. @ 9a2b3a1e: all MITM arms ≥ 0.70× on GHA median (33087088466, 33087091622, 33105885748 Linux; 33087085235, 33087088466, 33087091622 Windows). H2 plain→H2 plain Full 0.77–0.79× (Win) / 0.78× (Lin).

Client	Origin	Lite sustain	Full sustain	Lite÷Reverse	Full÷Reverse
HTTP/1 · plain	HTTP/1 · plain	35578 (88 MiB / 50.3% CPU)	34914 (97 MiB / 50.4% CPU)	0.97×	0.95×
HTTP/1 · plain	HTTP/1 · TLS	28326 (114 MiB / 49.8% CPU)	28227 (114 MiB / 49.8% CPU)	0.99×	0.99×
HTTP/1 · plain	HTTP/2 · plain	46274 (124 MiB / 54.5% CPU)	44157 (135 MiB / 54.7% CPU)	1.01×	0.97×
HTTP/1 · plain	HTTP/2 · TLS	37133 (147 MiB / 51% CPU)	36654 (146 MiB / 51.6% CPU)	0.98×	0.97×
HTTP/1 · plain	HTTP/3 · QUIC	26091 (142 MiB / 54.6% CPU)	25394 (145 MiB / 54.4% CPU)	1×	0.97×
HTTP/1 · TLS	HTTP/1 · plain	27963 (116 MiB / 50.4% CPU)	27273 (118 MiB / 50% CPU)	1.01×	0.98×
HTTP/1 · TLS	HTTP/1 · TLS	24331 (118 MiB / 49% CPU)	23589 (116 MiB / 48.9% CPU)	1.02×	0.99×
HTTP/1 · TLS	HTTP/2 · plain	33674 (144 MiB / 52.4% CPU)	32894 (144 MiB / 52% CPU)	0.99×	0.97×
HTTP/1 · TLS	HTTP/2 · TLS	29341 (163 MiB / 49.4% CPU)	29024 (160 MiB / 50.3% CPU)	0.99×	0.98×
HTTP/1 · TLS	HTTP/3 · QUIC	21874 (159 MiB / 53.2% CPU)	21237 (162 MiB / 53.3% CPU)	0.98×	0.95×
HTTP/2 · plain	HTTP/1 · plain	39834 (120 MiB / 54% CPU)	38577 (118 MiB / 53.7% CPU)	0.97×	0.93×
HTTP/2 · plain	HTTP/1 · TLS	32663 (135 MiB / 52.2% CPU)	31085 (123 MiB / 51.7% CPU)	1×	0.95×
HTTP/2 · plain	HTTP/2 · plain	73656 (93 MiB / 43.9% CPU)	71161 (97 MiB / 42.4% CPU)	0.76×	0.73×
HTTP/2 · plain	HTTP/2 · TLS	55089 (105 MiB / 41.2% CPU)	51723 (99 MiB / 40.6% CPU)	0.81×	0.76×
HTTP/2 · plain	HTTP/3 · QUIC	30054 (151 MiB / 51.4% CPU)	29788 (156 MiB / 51.7% CPU)	1×	0.99×
HTTP/2 · TLS	HTTP/1 · plain	36903 (123 MiB / 53.2% CPU)	35802 (127 MiB / 53.5% CPU)	0.95×	0.93×
HTTP/2 · TLS	HTTP/1 · TLS	30528 (130 MiB / 51.2% CPU)	29484 (124 MiB / 50.3% CPU)	1.01×	0.97×
HTTP/2 · TLS	HTTP/2 · plain	63418 (114 MiB / 43% CPU)	60875 (111 MiB / 42.8% CPU)	0.74×	0.71×
HTTP/2 · TLS	HTTP/2 · TLS	52660 (106 MiB / 40.8% CPU)	49767 (106 MiB / 39.5% CPU)	0.83×	0.79×
HTTP/2 · TLS	HTTP/3 · QUIC	28301 (166 MiB / 51.1% CPU)	27450 (154 MiB / 50.7% CPU)	1×	0.97×
HTTP/3 · QUIC	HTTP/1 · plain	22951 (150 MiB / 52.3% CPU)	22840 (150 MiB / 52.4% CPU)	0.97×	0.96×
HTTP/3 · QUIC	HTTP/1 · TLS	18924 (165 MiB / 50.9% CPU)	18029 (157 MiB / 49.7% CPU)	0.97×	0.92×
HTTP/3 · QUIC	HTTP/2 · plain	30229 (167 MiB / 57.6% CPU)	29277 (163 MiB / 56.6% CPU)	0.96×	0.93×
HTTP/3 · QUIC	HTTP/2 · TLS	26262 (163 MiB / 53.8% CPU)	25788 (168 MiB / 53.4% CPU)	0.95×	0.93×
HTTP/3 · QUIC	HTTP/3 · QUIC	20948 (163 MiB / 50.8% CPU)	20826 (162 MiB / 50.9% CPU)	0.96×	0.95×

macOS — Titanium vs nginx vs HAProxy vs Envoy vs YARP

Reverse

Median of 3 repeats on macos-15-intel (4-core / 14 GB). Bare reverse 5×5 @ 9a2b3a1e — compare-product 34441526151. Warmup 2s / measure 8s; concurrency 8, 16, 32, 64. Prefer TWP÷peer ratios over absolute RPS. RPS cells include median RSS / CPU at the peak-RPS step as <br><sub>(MiB / CPU%)</sub>. The RPS workflow installs Homebrew nginx (http_v3_module), Homebrew HAProxy with USE_QUIC (3.2 source fallback), Envoy (Homebrew bottle or pinned darwin-amd64 1.36.7), Homebrew libmsquic (+ DYLD_*), and YARP. Do not publish from macos-latest (3-core / 7 GB). Product 5×5 is ~56-byte JSON keep-alive GET; H2/H3 same-protocol cells are mostly header work with a tiny body (Titanium best case) — see Why this comparison is fair.

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
HTTP/1 · plain	HTTP/1 · plain	🥇 12131 (82 MiB / 32.8% CPU)	🥇 12131 (82 MiB / 32.8% CPU)	6938 (52 MiB / 11.6% CPU)	6938 (52 MiB / 11.6% CPU)	10945 (48 MiB / 23.2% CPU)	10945 (48 MiB / 23.2% CPU)	3332 (72 MiB / 22.4% CPU)	3332 (72 MiB / 22.4% CPU)	10805 (105 MiB / 33.6% CPU)	10805 (105 MiB / 33.6% CPU)
HTTP/1 · plain	HTTP/1 · TLS	🥇 11064 (93 MiB / 36.5% CPU)	🥇 11064 (93 MiB / 36.5% CPU)	5152 (73 MiB / 17.4% CPU)	5152 (73 MiB / 17.4% CPU)	7159 (53 MiB / 25.9% CPU)	7159 (53 MiB / 25.9% CPU)	0 (75 MiB / 21% CPU)	2011 (75 MiB / 21% CPU)	9146 (129 MiB / 36.8% CPU)	9146 (129 MiB / 36.8% CPU)
HTTP/1 · plain	HTTP/2 · plain	🥇 19185 (88 MiB / 33.8% CPU)	🥇 19185 (88 MiB / 33.8% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	11853 (50 MiB / 25.6% CPU)	11853 (50 MiB / 25.6% CPU)	6000 (73 MiB / 36.6% CPU)	6000 (73 MiB / 36.6% CPU)	18729 (111 MiB / 34% CPU)	18729 (111 MiB / 34% CPU)
HTTP/1 · plain	HTTP/2 · TLS	9927 (129 MiB / 32.9% CPU)	9927 (129 MiB / 32.9% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	8770 (51 MiB / 26.5% CPU)	8770 (51 MiB / 26.5% CPU)	5948 (73 MiB / 34.9% CPU)	5948 (73 MiB / 34.9% CPU)	🥇 13674 (116 MiB / 32.9% CPU)	🥇 13674 (116 MiB / 32.9% CPU)
HTTP/1 · plain	HTTP/3 · QUIC	3662 (90 MiB / 45.3% CPU)	3662 (90 MiB / 45.3% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not measured	Not measured	3393 (75 MiB / 38.2% CPU)	2631 (75 MiB / 38.2% CPU)	🥇 5774 (116 MiB / 33.4% CPU)	🥇 5774 (116 MiB / 33.4% CPU)
HTTP/1 · TLS	HTTP/1 · plain	🥇 14162 (94 MiB / 31.5% CPU)	🥇 14162 (94 MiB / 31.5% CPU)	8184 (74 MiB / 18.3% CPU)	8184 (74 MiB / 18.3% CPU)	9938 (64 MiB / 25.6% CPU)	9938 (64 MiB / 25.6% CPU)	4084 (79 MiB / 37.4% CPU)	4328 (79 MiB / 37.4% CPU)	7647 (115 MiB / 32.3% CPU)	7647 (115 MiB / 32.3% CPU)
HTTP/1 · TLS	HTTP/1 · TLS	6128 (97 MiB / 31.7% CPU)	6128 (97 MiB / 31.7% CPU)	4903 (81 MiB / 15.5% CPU)	4903 (81 MiB / 15.5% CPU)	🥇 9241 (67 MiB / 28% CPU)	🥇 9241 (67 MiB / 28% CPU)	3594 (80 MiB / 38.9% CPU)	3594 (80 MiB / 38.9% CPU)	6601 (172 MiB / 35% CPU)	6601 (172 MiB / 35% CPU)
HTTP/1 · TLS	HTTP/2 · plain	8317 (96 MiB / 32.8% CPU)	8317 (96 MiB / 32.8% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	🥇 12353 (66 MiB / 28% CPU)	🥇 12353 (66 MiB / 28% CPU)	5378 (80 MiB / 20.9% CPU)	3107 (80 MiB / 20.9% CPU)	10735 (122 MiB / 32.2% CPU)	10735 (122 MiB / 32.2% CPU)
HTTP/1 · TLS	HTTP/2 · TLS	8457 (161 MiB / 33.8% CPU)	8457 (161 MiB / 33.8% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	8558 (65 MiB / 29.3% CPU)	8558 (65 MiB / 29.3% CPU)	4938 (79 MiB / 39.8% CPU)	5195 (79 MiB / 39.8% CPU)	🥇 9023 (120 MiB / 28.7% CPU)	🥇 9023 (120 MiB / 28.7% CPU)
HTTP/1 · TLS	HTTP/3 · QUIC	2677 (112 MiB / 44.9% CPU)	2677 (112 MiB / 44.9% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not measured	Not measured	🥇 4392 (81 MiB / 37.8% CPU)	🥇 2176 (81 MiB / 37.8% CPU)	3589 (152 MiB / 31.7% CPU)	3589 (152 MiB / 31.7% CPU)
HTTP/2 · plain	HTTP/1 · plain	12398 (88 MiB / 35.1% CPU)	12398 (88 MiB / 35.1% CPU)	10718 (58 MiB / 15.4% CPU)	10718 (58 MiB / 15.4% CPU)	7841 (54 MiB / 18% CPU)	7841 (54 MiB / 18% CPU)	3391 (74 MiB / 20.7% CPU)	3391 (74 MiB / 20.7% CPU)	🥇 12934 (105 MiB / 37.8% CPU)	🥇 12934 (105 MiB / 37.8% CPU)
HTTP/2 · plain	HTTP/1 · TLS	13814 (94 MiB / 38.6% CPU)	13814 (94 MiB / 38.6% CPU)	12186 (85 MiB / 16% CPU)	12186 (85 MiB / 16% CPU)	6688 (56 MiB / 18.9% CPU)	6688 (56 MiB / 18.9% CPU)	5045 (77 MiB / 21.3% CPU)	5045 (77 MiB / 21.3% CPU)	🥇 14337 (121 MiB / 42.3% CPU)	🥇 14337 (121 MiB / 42.3% CPU)
HTTP/2 · plain	HTTP/2 · plain	🥇 30992 (73 MiB / 25.8% CPU)	🥇 30992 (73 MiB / 25.8% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	8615 (54 MiB / 17.5% CPU)	8615 (54 MiB / 17.5% CPU)	5442 (72 MiB / 20% CPU)	5442 (72 MiB / 20% CPU)	22898 (109 MiB / 37.5% CPU)	22898 (109 MiB / 37.5% CPU)
HTTP/2 · plain	HTTP/2 · TLS	🥇 24573 (77 MiB / 27.7% CPU)	🥇 24573 (77 MiB / 27.7% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	11876 (58 MiB / 19% CPU)	11876 (58 MiB / 19% CPU)	6393 (73 MiB / 20.2% CPU)	6393 (73 MiB / 20.2% CPU)	18700 (116 MiB / 34.6% CPU)	18700 (116 MiB / 34.6% CPU)
HTTP/2 · plain	HTTP/3 · QUIC	4840 (97 MiB / 49.3% CPU)	4840 (97 MiB / 49.3% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not measured	Not measured	3760 (74 MiB / 22.4% CPU)	3760 (74 MiB / 22.4% CPU)	🥇 8092 (117 MiB / 33.9% CPU)	🥇 8092 (117 MiB / 33.9% CPU)
HTTP/2 · TLS	HTTP/1 · plain	🥇 13829 (94 MiB / 39.2% CPU)	🥇 13829 (94 MiB / 39.2% CPU)	9070 (73 MiB / 14.5% CPU)	9070 (73 MiB / 14.5% CPU)	5237 (66 MiB / 16.1% CPU)	5237 (66 MiB / 16.1% CPU)	3516 (81 MiB / 20.4% CPU)	3516 (81 MiB / 20.4% CPU)	13572 (113 MiB / 37.3% CPU)	13572 (113 MiB / 37.3% CPU)
HTTP/2 · TLS	HTTP/1 · TLS	🥇 9182 (96 MiB / 37.3% CPU)	🥇 9182 (96 MiB / 37.3% CPU)	6812 (90 MiB / 15.3% CPU)	6812 (90 MiB / 15.3% CPU)	4920 (66 MiB / 18.2% CPU)	4920 (66 MiB / 18.2% CPU)	2505 (83 MiB / 20.1% CPU)	2505 (83 MiB / 20.1% CPU)	9080 (124 MiB / 40.6% CPU)	9080 (124 MiB / 40.6% CPU)
HTTP/2 · TLS	HTTP/2 · plain	🥇 47325 (81 MiB / 29% CPU)	🥇 47325 (81 MiB / 29% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	5750 (67 MiB / 5.9% CPU)	2851 (67 MiB / 5.9% CPU)	4704 (79 MiB / 19% CPU)	4704 (79 MiB / 19% CPU)	22095 (115 MiB / 37.6% CPU)	22095 (115 MiB / 37.6% CPU)
HTTP/2 · TLS	HTTP/2 · TLS	🥇 29038 (83 MiB / 27.1% CPU)	🥇 29038 (83 MiB / 27.1% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	6022 (67 MiB / 18.3% CPU)	6022 (67 MiB / 18.3% CPU)	4494 (79 MiB / 18.6% CPU)	4494 (79 MiB / 18.6% CPU)	16458 (118 MiB / 35% CPU)	16458 (118 MiB / 35% CPU)
HTTP/2 · TLS	HTTP/3 · QUIC	4226 (108 MiB / 36% CPU)	4226 (108 MiB / 36% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not measured	Not measured	1772 (81 MiB / 21.6% CPU)	1772 (81 MiB / 21.6% CPU)	🥇 4805 (125 MiB / 33.7% CPU)	🥇 4805 (125 MiB / 33.7% CPU)
HTTP/3 · QUIC	HTTP/1 · plain	5834 (101 MiB / 39.5% CPU)	5834 (101 MiB / 39.5% CPU)	0 (63 MiB / 11.2% CPU)	7464 (63 MiB / 11.2% CPU)	🥇 10727 (66 MiB / 22% CPU)	🥇 10727 (66 MiB / 22% CPU)	1595 (85 MiB / 26.4% CPU)	1595 (85 MiB / 26.4% CPU)	5558 (187 MiB / 35% CPU)	5558 (187 MiB / 35% CPU)
HTTP/3 · QUIC	HTTP/1 · TLS	3723 (119 MiB / 38.2% CPU)	3723 (119 MiB / 38.2% CPU)	0 (66 MiB / 11.1% CPU)	4316 (66 MiB / 11.1% CPU)	🥇 5430 (69 MiB / 21.8% CPU)	🥇 5430 (69 MiB / 21.8% CPU)	Not measured	Not measured	4786 (197 MiB / 35.5% CPU)	4786 (197 MiB / 35.5% CPU)
HTTP/3 · QUIC	HTTP/2 · plain	4576 (97 MiB / 41.3% CPU)	4576 (97 MiB / 41.3% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	🥇 5954 (66 MiB / 14.4% CPU)	🥇 5954 (66 MiB / 14.4% CPU)	Not measured	Not measured	5642 (176 MiB / 34.2% CPU)	5642 (176 MiB / 34.2% CPU)
HTTP/3 · QUIC	HTTP/2 · TLS	5532 (106 MiB / 41.4% CPU)	5532 (106 MiB / 41.4% CPU)	Not possible (no H2 upstream)	Not possible (no H2 upstream)	6633 (68 MiB / 21.5% CPU)	6633 (68 MiB / 21.5% CPU)	Not measured	Not measured	🥇 7854 (167 MiB / 32.5% CPU)	🥇 7854 (167 MiB / 32.5% CPU)
HTTP/3 · QUIC	HTTP/3 · QUIC	3460 (96 MiB / 35% CPU)	3460 (96 MiB / 35% CPU)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	Not measured	Not measured	Not measured	Not measured	🥇 4659 (188 MiB / 34.4% CPU)	🥇 4659 (188 MiB / 34.4% CPU)

MITM (TWP only)

Same Client×Origin wires with interception on (compare-product 34441526151). Lite = no-op handlers (unchanged-lite finish). Full = append-only header mutation (harness: one probe header each way; product: generic append-only relay via MitmCompressedRelayHelper). nginx/HAProxy/Envoy/YARP cannot MITM. Lite÷Reverse / Full÷Reverse vs bare reverse (same job / comparison-group shard). Completion gate: Lite ≥ 0.50× and Full ≥ 0.50× reverse sustain @ c=64 (median of 3 GHA runs); reverse TWP÷YARP ≥ 0.70× (no terminate-peer gate).

v1 append-only relay (2026-08-27): Pre-fix H2→H2 Full÷Reverse was 0.13–0.16× (32960766249). Post-fix @ df172718: H2 plain→H2 plain Full 0.77–0.79×, H3→H1 Full 0.91–0.93×, all MITM arms ≥ 0.70× on median of 33041445371, 33055267086, 33055272140.

v2 drop-only + non-unique append (2026-08-27): MitmStaticRebuildHelper rebuilds static HPACK/QPACK after 1–4 unique header drops; trailing non-unique appends stay on compressed relay. @ 9a2b3a1e: all MITM arms ≥ 0.70× on GHA median (33087088466, 33087091622, 33105885748 Linux; 33087085235, 33087088466, 33087091622 Windows). H2 plain→H2 plain Full 0.77–0.79× (Win) / 0.78× (Lin).

Client	Origin	Lite sustain	Full sustain	Lite÷Reverse	Full÷Reverse
HTTP/1 · plain	HTTP/1 · plain	11715 (84 MiB / 35.6% CPU)	11001 (84 MiB / 31.8% CPU)	0.97×	0.91×
HTTP/1 · plain	HTTP/1 · TLS	7193 (94 MiB / 31.1% CPU)	8678 (94 MiB / 35.3% CPU)	0.65×	0.78×
HTTP/1 · plain	HTTP/2 · plain	14866 (89 MiB / 36.3% CPU)	13716 (90 MiB / 35.7% CPU)	0.77×	0.71×
HTTP/1 · plain	HTTP/2 · TLS	11164 (109 MiB / 33.4% CPU)	9186 (108 MiB / 32.6% CPU)	1.12×	0.93×
HTTP/1 · plain	HTTP/3 · QUIC	3459 (91 MiB / 39.6% CPU)	3710 (90 MiB / 40.3% CPU)	0.94×	1.01×
HTTP/1 · TLS	HTTP/1 · plain	8052 (95 MiB / 30.3% CPU)	10123 (96 MiB / 33.1% CPU)	0.57×	0.71×
HTTP/1 · TLS	HTTP/1 · TLS	6914 (98 MiB / 36% CPU)	9190 (98 MiB / 33.3% CPU)	1.13×	1.5×
HTTP/1 · TLS	HTTP/2 · plain	8443 (123 MiB / 35.7% CPU)	8055 (102 MiB / 33.1% CPU)	1.02×	0.97×
HTTP/1 · TLS	HTTP/2 · TLS	8738 (133 MiB / 33.2% CPU)	7260 (157 MiB / 32.6% CPU)	1.03×	0.86×
HTTP/1 · TLS	HTTP/3 · QUIC	2633 (105 MiB / 42.6% CPU)	2842 (102 MiB / 46.3% CPU)	0.98×	1.06×
HTTP/2 · plain	HTTP/1 · plain	13742 (88 MiB / 38.4% CPU)	13895 (90 MiB / 37.5% CPU)	1.11×	1.12×
HTTP/2 · plain	HTTP/1 · TLS	14725 (96 MiB / 40.8% CPU)	10701 (96 MiB / 35.5% CPU)	1.07×	0.77×
HTTP/2 · plain	HTTP/2 · plain	24126 (76 MiB / 29.9% CPU)	24457 (76 MiB / 29.7% CPU)	0.78×	0.79×
HTTP/2 · plain	HTTP/2 · TLS	23664 (83 MiB / 31.7% CPU)	24893 (81 MiB / 31% CPU)	0.96×	1.01×
HTTP/2 · plain	HTTP/3 · QUIC	4822 (96 MiB / 45% CPU)	5035 (95 MiB / 45.9% CPU)	1×	1.04×
HTTP/2 · TLS	HTTP/1 · plain	13114 (92 MiB / 38.3% CPU)	12954 (91 MiB / 38.1% CPU)	0.95×	0.94×
HTTP/2 · TLS	HTTP/1 · TLS	9205 (97 MiB / 36% CPU)	9788 (97 MiB / 39.1% CPU)	1×	1.07×
HTTP/2 · TLS	HTTP/2 · plain	28081 (86 MiB / 32% CPU)	24916 (87 MiB / 30.4% CPU)	0.59×	0.53×
HTTP/2 · TLS	HTTP/2 · TLS	22874 (87 MiB / 29.6% CPU)	29489 (89 MiB / 32% CPU)	0.79×	1.02×
HTTP/2 · TLS	HTTP/3 · QUIC	4679 (105 MiB / 37.9% CPU)	4453 (106 MiB / 39.3% CPU)	1.11×	1.05×
HTTP/3 · QUIC	HTTP/1 · plain	4000 (102 MiB / 35% CPU)	4530 (102 MiB / 38.8% CPU)	0.69×	0.78×
HTTP/3 · QUIC	HTTP/1 · TLS	3931 (116 MiB / 38.3% CPU)	3992 (115 MiB / 40% CPU)	1.06×	1.07×
HTTP/3 · QUIC	HTTP/2 · plain	4272 (98 MiB / 39% CPU)	4718 (98 MiB / 39.8% CPU)	0.93×	1.03×
HTTP/3 · QUIC	HTTP/2 · TLS	5183 (106 MiB / 38.5% CPU)	4838 (108 MiB / 38.9% CPU)	0.94×	0.87×
HTTP/3 · QUIC	HTTP/3 · QUIC	3211 (94 MiB / 37.3% CPU)	3887 (95 MiB / 37.3% CPU)	0.93×	1.12×

Editions (CLI / Plus / Intercept)

Note: twp-reverse-http1 and other library rows use Core with probe-tuned settings (no logging, no Via header, probe-warmed certs). Edition rows use titanium run -c twp.yaml product defaults — prefer the ÷baseline ratio column over absolute RPS. Inspector GUI is not spawnable in the harness; session-path overhead is twp-cli-intercept-http1 (route RequestHeaderSet transform). Pre-origin Plus middleware (CIDR/WAF/JWT/rate-limit) runs on H1 terminate-lite without SessionEventArgs; JWT caches successful bearer validations. Maintainer gate thresholds live under Maintainer notes.

Median of 3 repeats @ 6d2a7c9d. Source: Actions 33259699099. Warmup 2s / measure 8s; concurrency 8–64; sustain = median peak RPS among SLO-pass steps @ c=64. RPS cells include (MiB / CPU%) at that step.

Arm	Win sustain	Win peak	Linux sustain	Linux peak	Win÷	Lin÷	Gate
twp-cli-reverse-http1 vs library	32214 (118 MiB / 43.4% CPU)	32219	47734 (151 MiB / 49.0% CPU)	47844	1.02×	1.04×	≥ 0.80×
twp-cli-reverse-http1-tls vs library TLS	26495 (138 MiB / 48.3% CPU)	26651	36932 (174 MiB / 48.5% CPU)	37200	1.01×	1.00×	≥ 0.80×
twp-cli-reverse-http1-route vs CLI	32394 (120 MiB / 47.7% CPU)	32428	47369 (149 MiB / 49.2% CPU)	47398	1.01×	0.99×	≥ 0.90×
twp-cli-plus-base-http1 vs CLI	32475 (123 MiB / 46.0% CPU)	32695	47322 (154 MiB / 49.8% CPU)	47744	1.01×	0.99×	≥ 0.90×
twp-cli-plus-cache-http1 (cold) vs CLI	34029 (118 MiB / 64.5% CPU)	34460	49409 (148 MiB / 62.3% CPU)	50104	1.06×	1.04×	≥ 0.70×
twp-cli-intercept-http1 vs CLI	25219 (124 MiB / 54.0% CPU)	25377	35969 (155 MiB / 51.2% CPU)	35977	0.78×	0.75×	≥ 0.70×
twp-cli-plus-waf-http1 vs CLI	31653 (124 MiB / 46.8% CPU)	31837	46367 (151 MiB / 49.6% CPU)	46685	0.98×	0.97×	≥ 0.80×
twp-cli-plus-cidr-http1 vs CLI	31730 (123 MiB / 51.1% CPU)	31757	46828 (150 MiB / 50.2% CPU)	47093	0.98×	0.98×	≥ 0.80×
twp-cli-plus-jwt-http1 vs CLI	30326 (138 MiB / 46.6% CPU)	30386	43906 (179 MiB / 49.5% CPU)	44433	0.94×	0.92×	≥ 0.70×
twp-cli-plus-ratelimit-http1 vs CLI	31713 (123 MiB / 48.8% CPU)	31944	46435 (151 MiB / 49.4% CPU)	46860	0.98×	0.97×	≥ 0.80×
twp-cli-plus-resilience-http1 vs CLI	32373 (128 MiB / 50.5% CPU)	32416	47302 (155 MiB / 49.5% CPU)	47654	1.00×	0.99×	≥ 0.85×
twp-cli-plus-discovery-file-http1 vs CLI	32209 (121 MiB / 46.5% CPU)	32332	47485 (151 MiB / 49.1% CPU)	47770	1.00×	0.99×	≥ 0.80×
twp-cli-plus-metrics-scrape-http1 vs CLI	32192 (126 MiB / 44.8% CPU)	32427	47348 (159 MiB / 49.5% CPU)	48665	1.00×	0.99×	≥ 0.80×
twp-cli-plus-cache-hit-http1 vs cache cold	34015 (117 MiB / 66.0% CPU)	34022	49153 (149 MiB / 63.1% CPU)	49358	1.00×	0.99×	≥ 0.90×
twp-cli-static-http1 vs CLI	53078 (109 MiB / 52.1% CPU)	53980	78333 (150 MiB / 48.4% CPU)	82955	1.65×	1.64×	≥ 0.85×
twp-cli-logging-http1 vs CLI	32440 (119 MiB / 47.4% CPU)	32581	47624 (150 MiB / 49.3% CPU)	48604	1.01×	1.00×	≥ 0.90×
twp-cli-lb-leasttime-http1 vs route	29649 (134 MiB / 54.7% CPU)	30386	44790 (180 MiB / 50.6% CPU)	44825	0.92×	0.95×	≥ 0.85×
twp-cli-dialect-twp-http1 vs CLI	32461 (114 MiB / 49.7% CPU)	32780	47898 (145 MiB / 49.3% CPU)	48154	1.01×	1.00×	≥ 0.90×

validate-edition-gates.ps1 passed on both OS for this run. Library baselines @ c=64 (same job): Win H1 31561 / TLS 26128; Linux H1 45818 / TLS 36869. Laptop smoke ratios stay on Performance Local Lab — Editions.

Cross-version (7.0 vs 6.0)

Same reverse matrix measured on Titanium 7.0 versus committed 6.0 baselines (33087088466). Median of 3 @ 0ef6d4dd. Source: Actions 33270571908. Sustain RPS @ c=64. Absolute 7.0÷6.0 ≈ 0.96–1.00× on same-protocol arms; peer-normalized ratios ≈ 0.90–1.03×.

Arm	Win 6.0	Win 7.0	Win÷	Linux 6.0	Linux 7.0	Lin÷
twp-reverse-http1	32525	32213	0.99×	32755	31932	0.97×
twp-reverse-http1-tls	26461	26570	1.00×	22784	22351	0.98×
twp-reverse-http2	76038	75316	0.99×	47690	46678	0.98×
twp-reverse-http2-cleartext	40347	40236	1.00×	34556	33316	0.96×
twp-reverse-http3	17529	17380	0.99×	19468	19423	1.00×
twp-reverse-http3-cleartext	19956	19529	0.98×	20395	20070	0.98×
yarp-reverse-http1 (peer)	27504	27254	0.99×	27713	27102	0.98×
yarp-reverse-http2 (peer)	35553	35189	0.99×	29111	28733	0.99×

Both OS CSVs passed the cross-version check for this run. MITM arms are measured with the product reverse matrix, not this reverse-only comparison.

Heavier reverse workloads

Same runners and harness as the tiny-GET tables, but with larger bodies, POST, lossy links, TLS cost, and architecture-sensitive paths (slow consumer / early response / duplex). Charts: rps-heavier-*-{windows,linux}.png. PUT with the same body is the same proxy work as POST; DELETE with no body matches GET — only POST is published. Bodies/POST/lossy stay half-duplex. Laptop numbers are on Performance Local Lab. How maintainers refresh these tables (modes, shards, paste scripts) is under Maintainer notes. Larger-body check: 64 / 256 KiB H2 TLS→H2 TLS is where body copy dominates headers — Titanium is behind YARP here (~0.69–0.76× at 64 KiB), unlike the tiny-GET H2↔H2 medals above.

Lossy link = userspace delay/drop shim (not kernel netem): TCP gets per-buffer delay + occasional whole-connection stalls (honest head-of-line for multiplexed HTTP/2); UDP gets per-datagram delay + drops (QUIC). Lossy tables publish HTTP/1, HTTP/2, and HTTP/3.

Windows — heavier reverse GET (64 KiB / 256 KiB)

Median of 3 repeats on windows-latest @ 9a2b3a1e. Source: Actions 34441570199 (compare-bodies). Warmup 2s / measure 8s. RPS cells include (MiB / CPU%) footprints.

Body	Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
64 KiB	HTTP/1 · TLS	HTTP/1 · plain	🥇 13,191 (122 MiB / 46.7% CPU)	13,191 (122 MiB / 46.7% CPU)	924 (142 MiB / 24.8% CPU)	924 (142 MiB / 24.8% CPU)	Not possible	Not possible	Not possible	Not possible	11,473 (133 MiB / 47.5% CPU)	11,473 (133 MiB / 47.5% CPU)
64 KiB	HTTP/2 · TLS	HTTP/1 · plain	🥇 11,769 (173 MiB / 46.2% CPU)	11,769 (173 MiB / 46.2% CPU)	898 (142 MiB / 24.8% CPU)	898 (142 MiB / 24.8% CPU)	Not possible	Not possible	Not possible	Not possible	9,560 (135 MiB / 48.1% CPU)	9,560 (135 MiB / 48.1% CPU)
64 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 5,958 (139 MiB / 41.5% CPU)	5,958 (139 MiB / 41.5% CPU)	Not possible (no QUIC)	Not possible (no QUIC)	Not possible	Not possible	Not possible	Not possible	5,060 (185 MiB / 51.0% CPU)	5,060 (185 MiB / 51.0% CPU)
256 KiB	HTTP/1 · TLS	HTTP/1 · plain	🥇 3,693 (136 MiB / 42.1% CPU)	3,693 (136 MiB / 42.1% CPU)	241 (142 MiB / 24.9% CPU)	241 (142 MiB / 24.9% CPU)	Not possible	Not possible	Not possible	Not possible	3,325 (130 MiB / 47.2% CPU)	3,325 (130 MiB / 47.2% CPU)
256 KiB	HTTP/2 · TLS	HTTP/1 · plain	🥇 3,784 (149 MiB / 36.2% CPU)	3,784 (149 MiB / 36.2% CPU)	230 (142 MiB / 24.8% CPU)	230 (142 MiB / 24.8% CPU)	Not possible	Not possible	Not possible	Not possible	2,625 (135 MiB / 45.2% CPU)	2,625 (135 MiB / 45.2% CPU)
256 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 1,535 (111 MiB / 40.4% CPU)	1,535 (111 MiB / 40.4% CPU)	Not possible (no QUIC)	Not possible (no QUIC)	Not possible	Not possible	Not possible	Not possible	1,385 (169 MiB / 44.6% CPU)	1,385 (169 MiB / 44.6% CPU)
64 KiB	HTTP/2 · plain	HTTP/1 · plain	🥇 14,010 (174 MiB / 40.5% CPU)	14,010 (174 MiB / 40.5% CPU)	5,865 (127 MiB / 24.0% CPU)	5,865 (127 MiB / 24.0% CPU)	Not possible	Not possible	Not possible	Not possible	12,443 (111 MiB / 39.6% CPU)	12,443 (111 MiB / 39.6% CPU)
64 KiB	HTTP/2 · TLS	HTTP/2 · plain	6,319 (79 MiB / 38.1% CPU)	6,319 (79 MiB / 38.1% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	🥇 9,493 (131 MiB / 50.3% CPU)	9,493 (131 MiB / 50.3% CPU)
64 KiB	HTTP/2 · TLS	HTTP/2 · TLS	5,660 (80 MiB / 36.4% CPU)	5,660 (80 MiB / 36.4% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	🥇 8,201 (141 MiB / 47.6% CPU)	8,201 (141 MiB / 47.6% CPU)
64 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	4,657 (129 MiB / 46.1% CPU)	4,657 (129 MiB / 46.1% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	🥇 5,021 (195 MiB / 47.6% CPU)	5,021 (195 MiB / 47.6% CPU)
64 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	🥇 6,149 (148 MiB / 42.2% CPU)	6,149 (148 MiB / 42.2% CPU)	Not possible (no QUIC)	Not possible (no QUIC)	Not possible	Not possible	Not possible	Not possible	4,496 (191 MiB / 49.0% CPU)	4,496 (191 MiB / 49.0% CPU)
256 KiB	HTTP/2 · plain	HTTP/1 · plain	🥇 4,457 (144 MiB / 31.8% CPU)	4,457 (144 MiB / 31.8% CPU)	1,759 (127 MiB / 23.6% CPU)	1,759 (127 MiB / 23.6% CPU)	Not possible	Not possible	Not possible	Not possible	3,719 (123 MiB / 37.8% CPU)	3,719 (123 MiB / 37.8% CPU)
256 KiB	HTTP/2 · TLS	HTTP/2 · plain	1,976 (84 MiB / 29.9% CPU)	1,976 (84 MiB / 29.9% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	🥇 2,325 (169 MiB / 46.5% CPU)	2,325 (169 MiB / 46.5% CPU)
256 KiB	HTTP/2 · TLS	HTTP/2 · TLS	1,758 (87 MiB / 29.7% CPU)	1,758 (87 MiB / 29.7% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	🥇 2,109 (153 MiB / 44.9% CPU)	2,109 (153 MiB / 44.9% CPU)
256 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	1,164 (153 MiB / 43.0% CPU)	1,164 (153 MiB / 43.0% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	🥇 1,319 (186 MiB / 44.8% CPU)	1,319 (186 MiB / 44.8% CPU)
256 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	🥇 1,404 (146 MiB / 41.0% CPU)	1,404 (146 MiB / 41.0% CPU)	Not possible (no QUIC)	Not possible (no QUIC)	Not possible	Not possible	Not possible	Not possible	1,269 (196 MiB / 44.8% CPU)	1,269 (196 MiB / 44.8% CPU)

nginx/Windows collapses on large reverse bodies in this harness; treat as same-OS only. H1 TLS 64 KiB ≈ 1.11× YARP; 256 KiB ≈ 1.23×. H2→H1 64 KiB ≈ 1.21×; H3→H1 64 KiB ≈ 1.18×.

Linux — heavier reverse GET (64 KiB / 256 KiB)

Median of 3 repeats @ 9a2b3a1e. Source: Actions 34441570199 (compare-bodies). Warmup 2s / measure 8s.

Body	Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
64 KiB	HTTP/1 · TLS	HTTP/1 · plain	8,010 (176 MiB / 45.3% CPU)	8,010 (176 MiB / 45.3% CPU)	5,555 (100 MiB / 52.4% CPU)	5,555 (100 MiB / 52.4% CPU)	🥇 9,012 (84 MiB / 40.6% CPU)	9,012 (84 MiB / 40.6% CPU)	7,524 (131 MiB / 45.8% CPU)	7,524 (131 MiB / 45.8% CPU)	6,504 (169 MiB / 48.4% CPU)	6,504 (169 MiB / 48.4% CPU)
64 KiB	HTTP/2 · TLS	HTTP/1 · plain	9,850 (231 MiB / 39.4% CPU)	9,850 (231 MiB / 39.4% CPU)	3,314 (103 MiB / 10.9% CPU)	3,314 (103 MiB / 10.9% CPU)	🥇 10,145 (87 MiB / 24.0% CPU)	10,145 (87 MiB / 24.0% CPU)	8,389 (141 MiB / 23.7% CPU)	8,389 (141 MiB / 23.7% CPU)	8,231 (164 MiB / 46.6% CPU)	8,231 (164 MiB / 46.6% CPU)
64 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 5,646 (184 MiB / 44.2% CPU)	5,646 (184 MiB / 44.2% CPU)	0 (129 MiB / 22.4% CPU)	1,678 (129 MiB / 22.4% CPU)	Not measured	Not measured	Not measured	Not measured	4,421 (231 MiB / 51.5% CPU)	4,421 (231 MiB / 51.5% CPU)
256 KiB	HTTP/1 · TLS	HTTP/1 · plain	2,766 (128 MiB / 37.7% CPU)	2,766 (128 MiB / 37.7% CPU)	1,735 (100 MiB / 53.6% CPU)	1,735 (100 MiB / 53.6% CPU)	🥇 2,974 (83 MiB / 34.0% CPU)	2,974 (83 MiB / 34.0% CPU)	2,701 (145 MiB / 33.9% CPU)	2,701 (145 MiB / 33.9% CPU)	2,164 (166 MiB / 45.8% CPU)	2,164 (166 MiB / 45.8% CPU)
256 KiB	HTTP/2 · TLS	HTTP/1 · plain	2,550 (209 MiB / 32.3% CPU)	2,550 (209 MiB / 32.3% CPU)	1,728 (103 MiB / 19.0% CPU)	1,728 (103 MiB / 19.0% CPU)	🥇 3,276 (86 MiB / 19.5% CPU)	3,276 (86 MiB / 19.5% CPU)	3,128 (168 MiB / 20.2% CPU)	3,128 (168 MiB / 20.2% CPU)	2,201 (163 MiB / 43.1% CPU)	2,201 (163 MiB / 43.1% CPU)
256 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 1,458 (159 MiB / 43.3% CPU)	1,458 (159 MiB / 43.3% CPU)	0 (120 MiB / 20.0% CPU)	24 (120 MiB / 20.0% CPU)	Not measured	Not measured	Not measured	Not measured	1,278 (218 MiB / 48.1% CPU)	1,278 (218 MiB / 48.1% CPU)
64 KiB	HTTP/2 · plain	HTTP/1 · plain	🥇 13,722 (251 MiB / 41.5% CPU)	13,722 (251 MiB / 41.5% CPU)	4,929 (80 MiB / 10.4% CPU)	4,929 (80 MiB / 10.4% CPU)	13,114 (72 MiB / 23.9% CPU)	13,114 (72 MiB / 23.9% CPU)	11,084 (133 MiB / 23.5% CPU)	11,084 (133 MiB / 23.5% CPU)	13,042 (143 MiB / 45.4% CPU)	13,042 (143 MiB / 45.4% CPU)
64 KiB	HTTP/2 · TLS	HTTP/2 · plain	3,419 (101 MiB / 40.6% CPU)	3,419 (101 MiB / 40.6% CPU)	Not possible	Not possible	0 (83 MiB / 19.9% CPU)	0 (83 MiB / 19.9% CPU)	0 (126 MiB / 22.4% CPU)	0 (126 MiB / 22.4% CPU)	🥇 4,578 (185 MiB / 46.3% CPU)	4,578 (185 MiB / 46.3% CPU)
64 KiB	HTTP/2 · TLS	HTTP/2 · TLS	4,682 (104 MiB / 37.3% CPU)	4,682 (104 MiB / 37.3% CPU)	Not possible	Not possible	3,969 (83 MiB / 24.5% CPU)	3,969 (83 MiB / 24.5% CPU)	🥇 6,266 (154 MiB / 23.0% CPU)	6,266 (154 MiB / 23.0% CPU)	6,174 (184 MiB / 46.0% CPU)	6,174 (184 MiB / 46.0% CPU)
64 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	2,941 (168 MiB / 52.7% CPU)	2,941 (168 MiB / 52.7% CPU)	Not possible	Not possible	Not measured	Not measured	Not measured	Not measured	🥇 3,040 (234 MiB / 48.9% CPU)	3,040 (234 MiB / 48.9% CPU)
64 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	🥇 6,442 (214 MiB / 41.6% CPU)	6,442 (214 MiB / 41.6% CPU)	0 (145 MiB / 17.0% CPU)	2,278 (145 MiB / 17.0% CPU)	Not measured	Not measured	Not measured	Not measured	4,761 (248 MiB / 49.0% CPU)	4,761 (248 MiB / 49.0% CPU)
256 KiB	HTTP/2 · plain	HTTP/1 · plain	3,918 (201 MiB / 33.0% CPU)	3,918 (201 MiB / 33.0% CPU)	1,946 (80 MiB / 14.2% CPU)	1,946 (80 MiB / 14.2% CPU)	4,300 (72 MiB / 18.6% CPU)	4,300 (72 MiB / 18.6% CPU)	🥇 4,479 (153 MiB / 21.0% CPU)	4,479 (153 MiB / 21.0% CPU)	3,945 (157 MiB / 38.1% CPU)	3,945 (157 MiB / 38.1% CPU)
256 KiB	HTTP/2 · TLS	HTTP/2 · plain	1,139 (111 MiB / 33.5% CPU)	1,139 (111 MiB / 33.5% CPU)	Not possible	Not possible	0 (81 MiB / 19.8% CPU)	0 (81 MiB / 19.8% CPU)	0 (126 MiB / 22.3% CPU)	0 (126 MiB / 22.3% CPU)	🥇 1,293 (186 MiB / 42.8% CPU)	1,293 (186 MiB / 42.8% CPU)
256 KiB	HTTP/2 · TLS	HTTP/2 · TLS	1,436 (114 MiB / 28.9% CPU)	1,436 (114 MiB / 28.9% CPU)	Not possible	Not possible	1,114 (89 MiB / 24.3% CPU)	1,114 (89 MiB / 24.3% CPU)	🥇 1,826 (160 MiB / 22.8% CPU)	1,826 (160 MiB / 22.8% CPU)	1,734 (185 MiB / 42.8% CPU)	1,734 (185 MiB / 42.8% CPU)
256 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	837 (187 MiB / 51.5% CPU)	837 (187 MiB / 51.5% CPU)	Not possible	Not possible	Not measured	Not measured	Not measured	Not measured	🥇 915 (246 MiB / 46.9% CPU)	915 (246 MiB / 46.9% CPU)
256 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	🥇 1,745 (189 MiB / 42.2% CPU)	1,745 (189 MiB / 42.2% CPU)	0 (143 MiB / 18.7% CPU)	619 (143 MiB / 18.7% CPU)	Not measured	Not measured	Not measured	Not measured	1,487 (249 MiB / 47.5% CPU)	1,487 (249 MiB / 47.5% CPU)

On this GHA pass TWP÷YARP H1 TLS ≈ 1.23× (64 KiB) / 1.28× (256 KiB); H2→H1 ≈ 1.23× / 1.16×; H3→H1 ≈ 1.27× / 1.13×. TWP÷nginx H1 TLS ≈ 1.43 / 1.58. Absolute RPS swings by VM; prefer ratios.

Windows — POST 64 KiB request + 64 KiB response

Median of 3 repeats on windows-latest @ 9a2b3a1e. Source: Actions 34441591377 (compare-post).

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
HTTP/1 · TLS	HTTP/1 · plain	🥇 5,883 (94 MiB / 46.6% CPU)	5,883 (94 MiB / 46.6% CPU)	352 (142 MiB / 24.7% CPU)	352 (142 MiB / 24.7% CPU)	Not possible	Not possible	Not possible	Not possible	4,093 (137 MiB / 55.7% CPU)	4,093 (137 MiB / 55.7% CPU)
HTTP/2 · TLS	HTTP/1 · plain	🥇 3,997 (184 MiB / 49.3% CPU)	3,997 (184 MiB / 49.3% CPU)	352 (143 MiB / 24.7% CPU)	352 (143 MiB / 24.7% CPU)	Not possible	Not possible	Not possible	Not possible	3,494 (134 MiB / 52.1% CPU)	3,494 (134 MiB / 52.1% CPU)
HTTP/3 · QUIC	HTTP/1 · plain	🥇 2,039 (162 MiB / 40.4% CPU)	2,039 (162 MiB / 40.4% CPU)	Not possible (no QUIC)	Not possible (no QUIC)	Not possible	Not possible	Not possible	Not possible	1,893 (203 MiB / 48.6% CPU)	1,893 (203 MiB / 48.6% CPU)
HTTP/2 · plain	HTTP/1 · plain	🥇 6,237 (182 MiB / 46.1% CPU)	6,237 (182 MiB / 46.1% CPU)	1,622 (130 MiB / 24.6% CPU)	1,622 (130 MiB / 24.6% CPU)	Not possible	Not possible	Not possible	Not possible	5,889 (125 MiB / 53.5% CPU)	5,889 (125 MiB / 53.5% CPU)
HTTP/2 · TLS	HTTP/2 · plain	8 (83 MiB / 0.2% CPU)	8 (83 MiB / 0.2% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	🥇 3,602 (143 MiB / 49.3% CPU)	3,602 (143 MiB / 49.3% CPU)
HTTP/2 · TLS	HTTP/2 · TLS	8 (88 MiB / 0.1% CPU)	8 (88 MiB / 0.1% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	🥇 2,837 (142 MiB / 46.8% CPU)	2,837 (142 MiB / 46.8% CPU)
HTTP/3 · QUIC	HTTP/2 · TLS	1,769 (178 MiB / 44.3% CPU)	1,769 (178 MiB / 44.3% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	🥇 1,857 (194 MiB / 47.4% CPU)	1,857 (194 MiB / 47.4% CPU)
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 1,935 (175 MiB / 42.5% CPU)	1,935 (175 MiB / 42.5% CPU)	Not possible (no QUIC)	Not possible (no QUIC)	Not possible	Not possible	Not possible	Not possible	1,768 (213 MiB / 47.8% CPU)	1,768 (213 MiB / 47.8% CPU)

TWP leads H1 POST (1.5× YARP), H2 POST (1.2× YARP), and H3 POST (~1.1× YARP).

Linux — POST 64 KiB request + 64 KiB response

Median of 3 repeats @ 9a2b3a1e. Source: Actions 34441591377 (compare-post).

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
HTTP/1 · TLS	HTTP/1 · plain	5,113 (133 MiB / 43.7% CPU)	5,113 (133 MiB / 43.7% CPU)	4,029 (100 MiB / 48.2% CPU)	4,029 (100 MiB / 48.2% CPU)	5,273 (86 MiB / 40.5% CPU)	5,273 (86 MiB / 40.5% CPU)	🥇 5,337 (131 MiB / 40.4% CPU)	5,337 (131 MiB / 40.4% CPU)	3,410 (176 MiB / 54.5% CPU)	3,410 (176 MiB / 54.5% CPU)
HTTP/2 · TLS	HTTP/1 · plain	🥇 3,356 (220 MiB / 47.1% CPU)	3,356 (220 MiB / 47.1% CPU)	1,563 (116 MiB / 21.2% CPU)	1,563 (116 MiB / 21.2% CPU)	2,012 (83 MiB / 24.0% CPU)	2,012 (83 MiB / 24.0% CPU)	0 (146 MiB / 20.5% CPU)	2,914 (146 MiB / 20.5% CPU)	2,767 (171 MiB / 47.9% CPU)	2,767 (171 MiB / 47.9% CPU)
HTTP/3 · QUIC	HTTP/1 · plain	🥇 3,052 (226 MiB / 43.7% CPU)	3,052 (226 MiB / 43.7% CPU)	557 (110 MiB / 25.0% CPU)	557 (110 MiB / 25.0% CPU)	Not measured	Not measured	Not measured	Not measured	2,778 (240 MiB / 49.2% CPU)	2,778 (240 MiB / 49.2% CPU)
HTTP/2 · plain	HTTP/1 · plain	🥇 5,779 (234 MiB / 42.8% CPU)	5,779 (234 MiB / 42.8% CPU)	2,488 (94 MiB / 22.3% CPU)	2,488 (94 MiB / 22.3% CPU)	3,011 (70 MiB / 23.1% CPU)	3,011 (70 MiB / 23.1% CPU)	0 (134 MiB / 22.7% CPU)	5,195 (134 MiB / 22.7% CPU)	4,169 (161 MiB / 41.4% CPU)	4,169 (161 MiB / 41.4% CPU)
HTTP/2 · TLS	HTTP/2 · plain	6 (118 MiB / 0.2% CPU)	6 (118 MiB / 0.2% CPU)	Not possible	Not possible	0 (90 MiB / 23.7% CPU)	0 (90 MiB / 23.7% CPU)	0 (153 MiB / 22.1% CPU)	0 (153 MiB / 22.1% CPU)	🥇 2,620 (179 MiB / 46.6% CPU)	2,620 (179 MiB / 46.6% CPU)
HTTP/2 · TLS	HTTP/2 · TLS	8 (115 MiB / 0.3% CPU)	8 (115 MiB / 0.3% CPU)	Not possible	Not possible	1,216 (88 MiB / 24.6% CPU)	1,216 (88 MiB / 24.6% CPU)	🥇 2,222 (147 MiB / 23.3% CPU)	2,222 (147 MiB / 23.3% CPU)	2,085 (181 MiB / 46.1% CPU)	2,085 (181 MiB / 46.1% CPU)
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 1,999 (236 MiB / 49.4% CPU)	1,999 (236 MiB / 49.4% CPU)	Not possible	Not possible	Not measured	Not measured	Not measured	Not measured	1,950 (245 MiB / 46.9% CPU)	1,950 (245 MiB / 46.9% CPU)
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 2,301 (253 MiB / 44.2% CPU)	2,301 (253 MiB / 44.2% CPU)	471 (120 MiB / 25.1% CPU)	471 (120 MiB / 25.1% CPU)	Not measured	Not measured	Not measured	Not measured	2,138 (274 MiB / 47.8% CPU)	2,138 (274 MiB / 47.8% CPU)

Linux nginx H1/H2/H3 POST completed (nginx.org mainline). TWP÷YARP H1 ≈ 1.5×; H2 ≈ 1.3×; H3 ≈ 1.1×. TWP÷nginx H3 ≈ 4×.

Windows — lossy / high-RTT (H2 HOL / H3 loss)

Userspace 5 ms one-way delay + 1% TCP connection stall (H1/H2) or UDP datagram drop (H3); 64 KiB GET. Median of 3 repeats on windows-latest @ 9a2b3a1e — 34441595456 (compare-lossy).

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
HTTP/1 · TLS	HTTP/1 · plain	🥇 663 (112 MiB / 3.0% CPU)	663 (112 MiB / 3.0% CPU)	652 (143 MiB / 16.4% CPU)	652 (143 MiB / 16.4% CPU)	Not possible	Not possible	Not possible	Not possible	662 (121 MiB / 4.5% CPU)	662 (121 MiB / 4.5% CPU)
HTTP/2 · TLS	HTTP/1 · plain	0 (120 MiB / 1.4% CPU)	86 (120 MiB / 1.4% CPU)	0 (142 MiB / 0.6% CPU)	17 (142 MiB / 0.6% CPU)	Not possible	Not possible	Not possible	Not possible	0 (99 MiB / 0.8% CPU)	16 (99 MiB / 0.8% CPU)
HTTP/3 · QUIC	HTTP/1 · plain	0 (67 MiB / 0.1% CPU)	0 (67 MiB / 0.1% CPU)	Not possible (no QUIC)	Not possible (no QUIC)	Not possible	Not possible	Not possible	Not possible	0 (80 MiB / 0.0% CPU)	0 (80 MiB / 0.0% CPU)
HTTP/2 · plain	HTTP/1 · plain	0 (130 MiB / 1.5% CPU)	89 (130 MiB / 1.5% CPU)	0 (128 MiB / 0.1% CPU)	17 (128 MiB / 0.1% CPU)	Not possible	Not possible	Not possible	Not possible	0 (91 MiB / 0.7% CPU)	16 (91 MiB / 0.7% CPU)
HTTP/2 · TLS	HTTP/2 · plain	0 (68 MiB / 0.6% CPU)	8 (68 MiB / 0.6% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	0 (100 MiB / 0.7% CPU)	16 (100 MiB / 0.7% CPU)
HTTP/2 · TLS	HTTP/2 · TLS	0 (71 MiB / 0.2% CPU)	8 (71 MiB / 0.2% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	0 (99 MiB / 0.8% CPU)	16 (99 MiB / 0.8% CPU)
HTTP/3 · QUIC	HTTP/2 · TLS	0 (70 MiB / 0.1% CPU)	0 (70 MiB / 0.1% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	0 (80 MiB / 0.0% CPU)	0 (80 MiB / 0.0% CPU)
HTTP/3 · QUIC	HTTP/1 · TLS	0 (69 MiB / 0.0% CPU)	0 (69 MiB / 0.0% CPU)	Not possible (no QUIC)	Not possible (no QUIC)	Not possible	Not possible	Not possible	Not possible	0 (78 MiB / 0.0% CPU)	0 (78 MiB / 0.0% CPU)

TWP H2 HOL leads (~3.31× YARP). H3 is the protocol-shape win vs H2 HOL on the same lossy session; Win H3 GHA remains 0 (laptop remeasure kept above).

Linux — lossy / high-RTT (H2 HOL / H3 loss)

Median of 3 repeats @ 9a2b3a1e. Source: 34441595456 (compare-lossy; lossy H3 uses quic-http3).

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
HTTP/1 · TLS	HTTP/1 · plain	1,195 (142 MiB / 13.7% CPU)	1,195 (142 MiB / 13.7% CPU)	1,205 (100 MiB / 12.6% CPU)	1,205 (100 MiB / 12.6% CPU)	🥇 1,207 (83 MiB / 7.8% CPU)	1,207 (83 MiB / 7.8% CPU)	1,193 (128 MiB / 9.3% CPU)	1,193 (128 MiB / 9.3% CPU)	1,194 (150 MiB / 17.5% CPU)	1,194 (150 MiB / 17.5% CPU)
HTTP/2 · TLS	HTTP/1 · plain	🥇 314 (182 MiB / 7.0% CPU)	314 (182 MiB / 7.0% CPU)	40 (99 MiB / 0.3% CPU)	40 (99 MiB / 0.3% CPU)	40 (86 MiB / 0.3% CPU)	40 (86 MiB / 0.3% CPU)	40 (135 MiB / 0.4% CPU)	40 (135 MiB / 0.4% CPU)	40 (127 MiB / 1.5% CPU)	40 (127 MiB / 1.5% CPU)
HTTP/3 · QUIC	HTTP/1 · plain	319 (151 MiB / 13.9% CPU)	319 (151 MiB / 13.9% CPU)	92 (109 MiB / 2.8% CPU)	92 (109 MiB / 2.8% CPU)	Not measured	Not measured	Not measured	Not measured	🥇 338 (177 MiB / 22.4% CPU)	338 (177 MiB / 22.4% CPU)
HTTP/2 · plain	HTTP/1 · plain	🥇 344 (174 MiB / 7.2% CPU)	344 (174 MiB / 7.2% CPU)	40 (78 MiB / 0.2% CPU)	40 (78 MiB / 0.2% CPU)	40 (69 MiB / 0.2% CPU)	40 (69 MiB / 0.2% CPU)	40 (129 MiB / 0.4% CPU)	40 (129 MiB / 0.4% CPU)	41 (121 MiB / 1.2% CPU)	41 (121 MiB / 1.2% CPU)
HTTP/2 · TLS	HTTP/2 · plain	0 (93 MiB / 0.6% CPU)	13 (93 MiB / 0.6% CPU)	Not possible	Not possible	0 (83 MiB / 2.7% CPU)	0 (83 MiB / 2.7% CPU)	0 (127 MiB / 6.3% CPU)	0 (127 MiB / 6.3% CPU)	🥇 40 (129 MiB / 1.6% CPU)	40 (129 MiB / 1.6% CPU)
HTTP/2 · TLS	HTTP/2 · TLS	0 (95 MiB / 0.7% CPU)	11 (95 MiB / 0.7% CPU)	Not possible	Not possible	🥇 42 (88 MiB / 0.6% CPU)	42 (88 MiB / 0.6% CPU)	40 (138 MiB / 0.5% CPU)	40 (138 MiB / 0.5% CPU)	40 (129 MiB / 1.6% CPU)	40 (129 MiB / 1.6% CPU)
HTTP/3 · QUIC	HTTP/2 · TLS	322 (140 MiB / 24.2% CPU)	322 (140 MiB / 24.2% CPU)	Not possible	Not possible	Not measured	Not measured	Not measured	Not measured	🥇 333 (183 MiB / 24.3% CPU)	333 (183 MiB / 24.3% CPU)
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 341 (161 MiB / 15.8% CPU)	341 (161 MiB / 15.8% CPU)	96 (113 MiB / 3.0% CPU)	96 (113 MiB / 3.0% CPU)	Not measured	Not measured	Not measured	Not measured	340 (185 MiB / 23.1% CPU)	340 (185 MiB / 23.1% CPU)

TWP H2 HOL ≫ YARP (~7.7×). H3 TWP÷YARP ≈ 1×.

Architecture-sensitive

These runs isolate slow app readers, origin-early response, HTTP/2 duplex, and WebSocket echo. See TWP vs YARP IO model. Laptop 1-rep numbers are on Performance Local Lab.

Median of 3 repeats on matched 4 vCPU / 16 GiB runners @ 9a2b3a1e (34441578556). Slow consumer = 256 KiB GET, 16 KiB read + 8 ms sleep. Early response = 64 KiB POST, origin writes after 8 KiB. Duplex HTTP/2 = overlapping 64 KiB POST on H2 TLS↔H2 TLS. WebSocket = echo round-trips/sec.

Lossy-link runs (slow network) are already published above; they are not a slow app reader.

Windows

Scenario	Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
Slow consumer (256 KiB GET, throttled client read)	HTTP/1 · TLS	HTTP/1 · plain	🥇 243 (101 MiB / 5.9% CPU)	243 (101 MiB / 5.9% CPU)	204 (143 MiB / 24.6% CPU)	204 (143 MiB / 24.6% CPU)	Not possible	Not possible	Not possible	Not possible	241 (109 MiB / 4.8% CPU)	241 (109 MiB / 4.8% CPU)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/1 · plain	256 (131 MiB / 3.7% CPU)	256 (131 MiB / 3.7% CPU)	230 (142 MiB / 24.4% CPU)	230 (142 MiB / 24.4% CPU)	Not possible	Not possible	Not possible	Not possible	🥇 256 (111 MiB / 5.4% CPU)	256 (111 MiB / 5.4% CPU)
Slow consumer (256 KiB GET, throttled client read)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 264 (101 MiB / 17.4% CPU)	264 (101 MiB / 17.4% CPU)	Not possible (no QUIC)	Not possible (no QUIC)	Not possible	Not possible	Not possible	Not possible	255 (158 MiB / 20.4% CPU)	255 (158 MiB / 20.4% CPU)
Early response (origin writes after first request chunk)	HTTP/1 · TLS	HTTP/1 · plain	🥇 5,611 (99 MiB / 45.5% CPU)	5,611 (99 MiB / 45.5% CPU)	363 (143 MiB / 24.7% CPU)	363 (143 MiB / 24.7% CPU)	Not possible	Not possible	Not possible	Not possible	3,919 (138 MiB / 53.9% CPU)	3,919 (138 MiB / 53.9% CPU)
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/1 · plain	🥇 3,934 (169 MiB / 48.8% CPU)	3,934 (169 MiB / 48.8% CPU)	0 (144 MiB / 24.8% CPU)	318 (144 MiB / 24.8% CPU)	Not possible	Not possible	Not possible	Not possible	3,200 (135 MiB / 53.4% CPU)	3,200 (135 MiB / 53.4% CPU)
Early response (origin writes after first request chunk)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 3,012 (152 MiB / 42.0% CPU)	3,012 (152 MiB / 42.0% CPU)	Not possible (no QUIC)	Not possible (no QUIC)	Not possible	Not possible	Not possible	Not possible	2,602 (203 MiB / 51.0% CPU)	2,602 (203 MiB / 51.0% CPU)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · plain	HTTP/1 · plain	248 (112 MiB / 3.6% CPU)	248 (112 MiB / 3.6% CPU)	248 (127 MiB / 9.4% CPU)	248 (127 MiB / 9.4% CPU)	Not possible	Not possible	Not possible	Not possible	🥇 256 (104 MiB / 6.0% CPU)	256 (104 MiB / 6.0% CPU)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/2 · TLS	0 (70 MiB / 1.0% CPU)	8 (70 MiB / 1.0% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	🥇 256 (136 MiB / 9.3% CPU)	256 (136 MiB / 9.3% CPU)
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/2 · TLS	0 (92 MiB / 0.1% CPU)	0 (92 MiB / 0.1% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	0 (110 MiB / 0.1% CPU)	0 (110 MiB / 0.1% CPU)
Duplex (both directions live)	HTTP/2 · TLS	HTTP/2 · TLS	0 (91 MiB / 0.1% CPU)	0 (91 MiB / 0.1% CPU)	Not possible	Not possible	Not possible	Not possible	Not possible	Not possible	0 (111 MiB / 0.1% CPU)	0 (111 MiB / 0.1% CPU)
Duplex (WebSocket / extended CONNECT)	HTTP/1 · TLS	HTTP/1 · plain	🥇 24,498 (97 MiB / 43.0% CPU)	24,498 (97 MiB / 43.0% CPU)	12,337 (143 MiB / 24.6% CPU)	12,337 (143 MiB / 24.6% CPU)	Not possible	Not possible	Not possible	Not possible	23,100 (89 MiB / 44.6% CPU)	23,100 (89 MiB / 44.6% CPU)

Linux

Scenario	Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
Slow consumer (256 KiB GET, throttled client read)	HTTP/1 · TLS	HTTP/1 · plain	469 (122 MiB / 9.7% CPU)	469 (122 MiB / 9.7% CPU)	411 (100 MiB / 9.9% CPU)	411 (100 MiB / 9.9% CPU)	466 (83 MiB / 6.0% CPU)	466 (83 MiB / 6.0% CPU)	🥇 470 (140 MiB / 6.2% CPU)	470 (140 MiB / 6.2% CPU)	411 (146 MiB / 14.2% CPU)	411 (146 MiB / 14.2% CPU)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/1 · plain	🥇 480 (149 MiB / 13.1% CPU)	480 (149 MiB / 13.1% CPU)	374 (102 MiB / 5.0% CPU)	374 (102 MiB / 5.0% CPU)	475 (85 MiB / 4.3% CPU)	475 (85 MiB / 4.3% CPU)	477 (149 MiB / 4.5% CPU)	477 (149 MiB / 4.5% CPU)	477 (146 MiB / 15.0% CPU)	477 (146 MiB / 15.0% CPU)
Slow consumer (256 KiB GET, throttled client read)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 473 (131 MiB / 34.1% CPU)	473 (131 MiB / 34.1% CPU)	0 (119 MiB / 21.9% CPU)	122 (119 MiB / 21.9% CPU)	Not measured	Not measured	Not measured	Not measured	468 (198 MiB / 39.8% CPU)	468 (198 MiB / 39.8% CPU)
Early response (origin writes after first request chunk)	HTTP/1 · TLS	HTTP/1 · plain	4,671 (132 MiB / 47.4% CPU)	4,671 (132 MiB / 47.4% CPU)	0 (100 MiB / 49.6% CPU)	3,397 (100 MiB / 49.6% CPU)	🥇 4,886 (84 MiB / 43.3% CPU)	4,886 (84 MiB / 43.3% CPU)	0 (129 MiB / 25.6% CPU)	2,584 (129 MiB / 25.6% CPU)	3,159 (176 MiB / 56.3% CPU)	3,159 (176 MiB / 56.3% CPU)
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/1 · plain	🥇 3,266 (242 MiB / 48.3% CPU)	3,266 (242 MiB / 48.3% CPU)	0 (113 MiB / 24.7% CPU)	1,365 (113 MiB / 24.7% CPU)	2,480 (85 MiB / 24.6% CPU)	2,480 (85 MiB / 24.6% CPU)	0 (148 MiB / 23.9% CPU)	2,653 (148 MiB / 23.9% CPU)	2,198 (169 MiB / 48.3% CPU)	2,198 (169 MiB / 48.3% CPU)
Early response (origin writes after first request chunk)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 4,504 (218 MiB / 43.2% CPU)	4,504 (218 MiB / 43.2% CPU)	0 (125 MiB / 22.7% CPU)	1,090 (125 MiB / 22.7% CPU)	Not measured	Not measured	Not measured	Not measured	3,217 (274 MiB / 48.7% CPU)	3,217 (274 MiB / 48.7% CPU)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · plain	HTTP/1 · plain	🥇 478 (136 MiB / 15.5% CPU)	478 (136 MiB / 15.5% CPU)	360 (79 MiB / 8.6% CPU)	360 (79 MiB / 8.6% CPU)	473 (69 MiB / 6.4% CPU)	473 (69 MiB / 6.4% CPU)	473 (147 MiB / 4.7% CPU)	473 (147 MiB / 4.7% CPU)	475 (150 MiB / 16.2% CPU)	475 (150 MiB / 16.2% CPU)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/2 · TLS	8 (97 MiB / 2.5% CPU)	8 (97 MiB / 2.5% CPU)	Not possible	Not possible	455 (89 MiB / 22.6% CPU)	455 (89 MiB / 22.6% CPU)	466 (152 MiB / 12.9% CPU)	466 (152 MiB / 12.9% CPU)	🥇 466 (168 MiB / 31.1% CPU)	466 (168 MiB / 31.1% CPU)
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/2 · TLS	0 (112 MiB / 0.1% CPU)	0 (112 MiB / 0.1% CPU)	Not possible	Not possible	Not measured	Not measured	0 (156 MiB / 19.5% CPU)	3,680 (156 MiB / 19.5% CPU)	0 (146 MiB / 0.2% CPU)	0 (146 MiB / 0.2% CPU)
Duplex (both directions live)	HTTP/2 · TLS	HTTP/2 · TLS	0 (112 MiB / 0.1% CPU)	0 (112 MiB / 0.1% CPU)	Not possible	Not possible	0 (91 MiB / 0.1% CPU)	0 (91 MiB / 0.1% CPU)	0 (151 MiB / 19.4% CPU)	3,724 (151 MiB / 19.4% CPU)	0 (151 MiB / 0.1% CPU)	0 (151 MiB / 0.1% CPU)
Duplex (WebSocket / extended CONNECT)	HTTP/1 · TLS	HTTP/1 · plain	28,567 (125 MiB / 44.0% CPU)	28,567 (125 MiB / 44.0% CPU)	🥇 33,775 (100 MiB / 35.5% CPU)	33,775 (100 MiB / 35.5% CPU)	31,870 (84 MiB / 39.3% CPU)	31,870 (84 MiB / 39.3% CPU)	31,525 (127 MiB / 39.4% CPU)	31,525 (127 MiB / 39.4% CPU)	27,109 (125 MiB / 44.1% CPU)	27,109 (125 MiB / 44.1% CPU)

Slow consumer is sleep-bound; H1/H2/H3 sit in the same band. Early-response H1/H2/H3: TWP leads (H1 early ≈ 2.00× / 1.47× YARP Win/Linux). Duplex H2: YARP leads by design — Win ≈ 0.59× (1,270 / 2,135), Linux ≈ 0.15× (282 / 1,882); irreducible concurrent-copier cell (see IO model). WebSocket: TWP÷YARP Windows ≈ 1.06×; Linux nginx leads.

TLS termination cost (H1 TLS → cleartext origin)

Isolates keep-alive tiny GET vs new connection per request (handshake-dominated) vs keep-alive 256 KiB. Product comparison uses RPS and end-to-end latency; TWP can also capture ClientTlsTiming when TWP_RPS_CAPTURE_TLS=1 (child process) — nginx/YARP have no equivalent hook.

Windows

Median of 3 repeats on windows-latest @ 9a2b3a1e. Source: Actions 34441599658. Absolute RPS on GHA swings hard; prefer TWP÷YARP.

Workload	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
Keep-alive · tiny GET	🥇 20,612 (87 MiB / 47.9% CPU)	20,612 (87 MiB / 47.9% CPU)	8,972 (142 MiB / 24.8% CPU)	8,972 (142 MiB / 24.8% CPU)	Not possible	Not possible	Not possible	Not possible	18,392 (105 MiB / 50.9% CPU)	18,392 (105 MiB / 50.9% CPU)
New-connection · tiny GET	🥇 732 (88 MiB / 9.5% CPU)	732 (88 MiB / 9.5% CPU)	0 (140 MiB / 24.4% CPU)	248 (140 MiB / 24.4% CPU)	Not possible	Not possible	Not possible	Not possible	725 (113 MiB / 10.7% CPU)	725 (113 MiB / 10.7% CPU)
Keep-alive · 256 KiB GET	🥇 2,741 (130 MiB / 47.0% CPU)	2,741 (130 MiB / 47.0% CPU)	0 (142 MiB / 24.6% CPU)	162 (142 MiB / 24.6% CPU)	Not possible	Not possible	Not possible	Not possible	2,699 (136 MiB / 49.4% CPU)	2,699 (136 MiB / 49.4% CPU)

Linux

Median of 3 repeats @ 9a2b3a1e. Source: Actions 34441599658.

Workload	TWP sustain	TWP peak	nginx sustain	nginx peak	HAProxy sustain	HAProxy peak	Envoy sustain	Envoy peak	YARP sustain	YARP peak
Keep-alive · tiny GET	36,114 (108 MiB / 49.4% CPU)	36,114 (108 MiB / 49.4% CPU)	🥇 44,372 (102 MiB / 40.8% CPU)	44,372 (102 MiB / 40.8% CPU)	43,256 (84 MiB / 43.3% CPU)	43,256 (84 MiB / 43.3% CPU)	27,636 (128 MiB / 57.3% CPU)	27,636 (128 MiB / 57.3% CPU)	32,712 (135 MiB / 49.8% CPU)	32,712 (135 MiB / 49.8% CPU)
New-connection · tiny GET	1,549 (128 MiB / 40.0% CPU)	1,549 (128 MiB / 40.0% CPU)	🥇 1,598 (103 MiB / 36.2% CPU)	1,598 (103 MiB / 36.2% CPU)	1,454 (85 MiB / 37.7% CPU)	1,454 (85 MiB / 37.7% CPU)	1,379 (129 MiB / 44.5% CPU)	1,379 (129 MiB / 44.5% CPU)	1,525 (149 MiB / 38.9% CPU)	1,525 (149 MiB / 38.9% CPU)
Keep-alive · 256 KiB GET	3,790 (126 MiB / 31.8% CPU)	3,790 (126 MiB / 31.8% CPU)	2,421 (101 MiB / 48.7% CPU)	2,421 (101 MiB / 48.7% CPU)	🥇 3,940 (84 MiB / 28.4% CPU)	3,940 (84 MiB / 28.4% CPU)	3,544 (145 MiB / 32.3% CPU)	3,544 (145 MiB / 32.3% CPU)	3,031 (160 MiB / 42.1% CPU)	3,031 (160 MiB / 42.1% CPU)

All three workloads are >1.00× YARP on both OS. nginx leads Linux keep-alive tiny and Linux new-connection; TWP is second, YARP third.

Unary gRPC (H2 TLS)

Unary Echo RPC/s @ c=64 over H2 TLS→H2 TLS for Titanium, YARP, nginx (grpc_pass), HAProxy, and Envoy (OS-possible peers only). Not folded into the architecture-sensitive tables. Median of 3 repeats @ 9a2b3a1e — 34441548073.

OS	Titanium	YARP	nginx	HAProxy	Envoy
Windows	53,762 (88 MiB / 23.6% CPU)	30,754 (123 MiB / 46.9% CPU)	Not measured	Not possible	Not possible
Linux	42,675 (120 MiB / 30.3% CPU)	24,232 (159 MiB / 41.6% CPU)	Not measured	8,683 (84 MiB / 24.6% CPU)	16,110 (128 MiB / 20.6% CPU)
macOS	14,924 (94 MiB / 20.1% CPU)	8,581 (128 MiB / 28.4% CPU)	Not measured	Not measured	Not measured

Other measurements

What	Result
HTTPS TTFB vs direct (median, 14 hosts)	Cold ≈ parity (−1 ms); warm −25 ms (proxy faster)
HTTP/1 loopback GET (no body intercept)	~128 µs, ~9.9 KB allocated / request
Basic example footprint (Release, after load)	~74 MB working set · ~24–29 MB private bytes

dotnet run -c Release --project benchmarks/Titanium.Web.Proxy.Benchmarks -- --filter '*Throughput*'

Local laptop BDN @ 9d7c2966 (Release):

Benchmark	Setup	Mean	Allocated / op
HTTP/1 GET through proxy	Passthrough	128 µs	9.9 KB
HTTP/2 multiplexed GETs	10 concurrent streams	253 µs / batch	~4.4 KB / request

Raising limits on large hosts

There is no artificial upper clamp on server defaults. Per-endpoint overrides:

Knob	Scope	Default	Override
ProxyServer.MaxCachedConnections	process, per upstream host	128	any ≥ 1
ProxyEndPoint.MaxCachedConnections	endpoint → pool depth for that EP’s sessions	null (use server)	e.g. 256 on reverse EP
ProxyEndPoint.MaxConcurrentClients	endpoint admission	null (off)	any ≥ 1
ResourceLimits.MaxConcurrentStreamsPerConnection	H2 streams	256	ProxyResourceLimits.Create(...)
TransparentQuicProxyEndPoint.MaxInboundBidirectionalStreams	H3	100 (probe uses 256)	property on EP
ForwardCleartext	transparent TLS terminate	false	true + decrypt

proxy.MaxCachedConnections = 512;
proxy.ResourceLimits = ProxyResourceLimits.Create(
    /* … */,
    maxConcurrentStreamsPerConnection: 1000,
    maxCachedConnectionsPerHost: 512,
    /* … */);

var ep = new TransparentProxyEndPoint(IPAddress.Any, 443, decryptSsl: true)
{
    ForwardHost = "127.0.0.1",
    ForwardPort = 8080,
    ForwardCleartext = true,
    MaxCachedConnections = 256,
    GenericCertificateName = "example.com"
};
ep.BeforeSslAuthenticate += (_, a) =>
{
    a.UpstreamHttpProtocol = UpstreamHttpProtocol.Http11;
    a.AllowHttpProtocolTranslation = true; // HTTP/2 client → HTTP/1 origin
    return Task.CompletedTask;
};

Maintainer notes

CI cadence, shards, paste scripts, and gate floors for people refreshing these tables. Consumers can skip this section.

Tiered cadence

Tier	Mode	When
Daily / per-PR	compare-spot	minutes
Milestone	compare-terminate / compare-matrix	~1–2h
Editions	compare-editions	~60 min (CLI / Plus / Intercept stress arms)
Cross-version (Gate 2)	compare-cross-version	~1–2h vs committed 6.0 baselines
Pre-wiki smoke	compare-product-smoke (Linux 2 shards, repeats=1)	~30–60 min; required before full product
Release / wiki	compare-product (3 comparison-group shards × Win/Linux/mac)	~2–2½h wall (Free account queues beyond 20 jobs)
Unary gRPC	compare-grpc	Win/Linux/mac; RPC/s @ c=64
Heavier tables	compare-bodies (2 shards) / post / lossy / arch (3 shards) / tls-cost	dispatch independently

See PERF-GATES.md.

Arm shards (comparison groups)

--arm-shard i/n (workflow input arm_shard) partitions wiki rows (Client×Origin + heavier/arch workload suffix), not individual proxy arms. Paste unions shard CSVs (paste-compare-product-wiki.ps1 -RunIds …; heavier RUNS lists). Table “Source: Actions” may list several run URLs for one table — do not mix SHAs. Free GitHub accounts allow 20 concurrent jobs (~34 for a full suite); extras queue, they do not fail. Dispatch product (and cross-version) first when wall clock matters.

One wiki row = one GHA job’s Client×Origin cell set: TWP + YARP + nginx + HAProxy + Envoy (when the OS can run them) + TWP Lite + Full stay on the same VM. Shards split rows, not individual proxies — so TWP÷YARP and Lite÷Reverse remain same-job ratios. Do not compare absolute RPS across shards (different VMs).

Gate thresholds

• Reverse product signal: TWP÷YARP ≥ 0.70 (nginx / HAProxy / Envoy are wiki and chart peers only — no CI gate).
• MITM overhead: Lite÷Reverse ≥ 0.50 and Full÷Reverse ≥ 0.50 (median of 3 GHA runs @ c=64). Absolute RPS moves with runner heat; ratios are the claim.
• Editions: see PERF-GATES.md and validate-edition-gates.ps1.
• Cross-version (Gate 2) before a major tag: run compare-cross-version (reverse matrix, routes unset) on develop vs committed 6.0 baselines (baseline-6.0-win.csv / baseline-6.0-linux.csv).

Gate	Threshold
Peer-normalized RPS (when YARP peer exists)	(TWP÷YARP)_7 ÷ (TWP÷YARP)_6 ≥ 0.90 or current (TWP÷YARP) ≥ 0.90
Absolute RPS floor (TWP arms)	7.0 ÷ 6.0 ≥ 0.70 (runner heat — peers move with the box)
RSS	7.0 ÷ 6.0 ≤ 1.20 per TWP arm @ c=64

nginx/YARP absolute RPS is not gated. See validate-cross-version.ps1.

pwsh tools/RpsLoadProbe/run-rps.ps1 -Mode compare-cross-version
pwsh tools/RpsLoadProbe/validate-cross-version.ps1 `
  -BaselineCsv tools/RpsLoadProbe/results/baseline-6.0-win.csv `
  -CurrentCsv  tools/RpsLoadProbe/results/rps-ramp-*.csv

Paste / regenerate

Product refresh and heavier tables:

pwsh tools/RpsLoadProbe/run-rps.ps1 -Mode compare-product
pwsh tools/RpsLoadProbe/run-rps.ps1 -Mode compare-editions
pwsh tools/RpsLoadProbe/validate-edition-gates.ps1 -CsvPath tools/RpsLoadProbe/results/rps-ramp-*.csv
pwsh tools/RpsLoadProbe/run-rps.ps1 -Mode compare-bodies
pwsh tools/RpsLoadProbe/run-rps.ps1 -Mode compare-post
pwsh tools/RpsLoadProbe/run-rps.ps1 -Mode compare-lossy
pwsh tools/RpsLoadProbe/run-rps.ps1 -Mode compare-tls-cost
pwsh tools/RpsLoadProbe/run-rps.ps1 -Mode compare-arch
pwsh tools/RpsLoadProbe/run-rps.ps1 -Mode compare-saturation

Harness details: RpsLoadProbe. The harness does not disable TWP safety that peers also skip, and it does not retune to pass gates (PERF-GATES.md). MITM leaf keys are a shared RSA/ECDSA pair per CertificateManager (intentional RPS tradeoff).