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 tables use RPC/s). Numbers are Release builds on matched GitHub-hosted runners: Windows and Linux at 4 vCPU / 16 GiB, macOS at macos-15 Apple Silicon 3-core / 7 GB. We compare products within an OS, not OS against OS, so read within one 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). 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 — Not possible on Windows, so their columns are omitted there.
• MITM (HTTPS decryption with forged certificates) is Titanium-only; peers cannot MITM. The published MITM table is Linux only. It shows 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.
• HTTP/2 and HTTP/3 spread that concurrency across min(concurrency, max(4 × cores, 16)) client connections, the same connection count h2load and wrk fix with -c. On these 4 vCPU runners that is 8, 16, 16, and 16 connections at c = 8, 16, 32, 64.
• 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 Win/Linux tiny GET). With larger bodies, see Heavier reverse (@ 482f5984).

How to read the tables

• Bold RPS is sustain (last concurrency that still met error/latency SLOs). When peak differs, it appears in the same cell as <sub>(peak N · …)</sub> with RSS/CPU. When sustain equals peak, peak is omitted. Footnotes also include microseconds of CPU per request when CPU was sampled.
• 🥇 = best among OS-possible peers on that row (highest sustain RPS; on a tie, lower memory then lower CPU%). On 64 KiB and 256 KiB rows the medal follows goodput (sustain RPS × request and response bytes). On tiny Reverse that is Titanium / nginx / YARP (plus HAProxy / Envoy on Linux/macOS). The same rule applies to heavier reverse, architecture-sensitive, TLS-cost, gRPC, and WebSocket peer tables. Gold on tiny GET is that frame-copy best case, not “Titanium is 1.7× on all reverse.” For larger-body H2→H2, see the heavier tables.
• MITM is published on Linux only (Titanium-only, no peer medal). 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. When every row would be Not possible for a peer, that column is omitted and a note above the table explains why.

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, Apple Silicon)
  • 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)
• Heavier reverse workloads
• Unary gRPC (H2 TLS)
• Unary gRPC (H2 TLS → h2c)
• WebSocket (H1 TLS → H1 TLS)
• WebSocket (H2 TLS 8441 → H1)
• Other measurements
• Raising limits on large hosts
• Maintainer notes

Measurement environment

All three OS use public-repo GitHub-hosted runners: Windows / Linux at 4 vCPU / 16 GiB / 14 GB SSD, macOS at macos-15 Apple Silicon (M1) 3-core / 7 GB (pinned, not macos-latest). macOS is Apple Silicon because that is what Mac users run today; the runner sizes differ, which is fine because we compare products within an OS. 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.

HTTP/2 and HTTP/3 open min(concurrency, max(4 × cores, 16)) client connections, one SocketsHttpHandler each. On the 4 vCPU runners that is 8, 16, 16, and 16 connections at c = 8, 16, 32, 64. HTTP/1 stays one connection per in-flight request. The generator is in-process .NET SocketsHttpHandler on the same machine as the origin and the proxy. Client, origin, and proxy share those cores, so the proxy's CPU share stays below a full box.

nginx TCP and TLS listen lines use reuseport. Linux and Windows keep worker_processes auto. macOS stays worker_processes 1. HAProxy nbthread and Envoy --concurrency stay at the processor count.

Large-body flow control is the same class on every peer: nginx http2_chunk_size 16k; HAProxy tune.bufsize 65536 and tune.h2.initial-window-size 65536; Envoy stream window 65535 bytes and connection window 1 MiB. Titanium's defaults stay a 65535-byte connection window and a 16384-byte max frame.

Cells for 64 KiB and 256 KiB bodies include goodput (MiB/s) and microseconds of CPU per request. Those rows are ranked on goodput. Smaller bodies stay ranked on RPS.

On the Linux H2 TLS→H1 plain row at c=64 (16 connections, @ 482f5984), proxy CPU was about 44% for nginx, 46% for HAProxy, and 63% for Envoy. Windows nginx on the same H2 client shape stayed near 25% (one core). That measured shape is the one published here.

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, Apple Silicon)

OS	macOS 15 (GitHub-hosted macos-15, Apple Silicon arm64)
CPU	3 logical processors (Apple M1)
RAM	7 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-arm64 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

Use the pinned macos-15 label (not macos-latest) so the image does not change between runs.

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 @ 482f5984 — 37210519205. 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	RPS	% of origin-HttpClient
origin-direct	dotnet-httpclient	51,094 (55 MiB / 40.5% CPU · 31.7 µs CPU/req)	100.0%
origin-direct-bombardier	bombardier	38,568 (56 MiB / 21.2% CPU · 21.9 µs CPU/req)	75.5%
bare-reverse-http1	dotnet-httpclient	25,550 (55 MiB / 43.1% CPU · 67.6 µs CPU/req)	50.0%
nginx-reverse-http1	dotnet-httpclient	13,356 (126 MiB / 24.8% CPU · 74.2 µs CPU/req)	26.1%
yarp-reverse-http1	dotnet-httpclient	21,546 (88 MiB / 49.4% CPU · 91.7 µs CPU/req)	42.2%
twp-reverse-http1	dotnet-httpclient	🥇 25,433 (72 MiB / 47.2% CPU · 74.3 µs CPU/req)	49.8%

Linux (ubuntu-latest)

Arm	Generator	RPS	% of origin-HttpClient
origin-direct	dotnet-httpclient	70,838 (80 MiB / 42.9% CPU · 24.2 µs CPU/req)	100.0%
origin-direct-bombardier	bombardier	41,999 (80 MiB / 34.9% CPU · 33.2 µs CPU/req)	59.3%
bare-reverse-http1	dotnet-httpclient	32,377 (67 MiB / 44.4% CPU · 54.8 µs CPU/req)	45.7%
nginx-reverse-http1	dotnet-httpclient	🥇 39,318 (76 MiB / 41.3% CPU · 42.0 µs CPU/req)	55.5%
yarp-reverse-http1	dotnet-httpclient	27,756 (115 MiB / 50.0% CPU · 72.1 µs CPU/req)	39.2%
twp-reverse-http1	dotnet-httpclient	31,478 (83 MiB / 49.5% CPU · 62.9 µs CPU/req)	44.4%

macOS (macos-15, Apple Silicon)

Arm	Generator	RPS	% of origin-HttpClient
origin-direct	dotnet-httpclient	85,921 (90 MiB / 34.0% CPU · 15.8 µs CPU/req)	100.0%
origin-direct-bombardier	bombardier	63,764 (89 MiB / 24.6% CPU · 15.4 µs CPU/req)	74.2%
bare-reverse-http1	dotnet-httpclient	33,705 (92 MiB / 32.5% CPU · 38.6 µs CPU/req)	39.2%
nginx-reverse-http1	dotnet-httpclient	26,475 (68 MiB / 21.2% CPU · 32.1 µs CPU/req)	30.8%
yarp-reverse-http1	dotnet-httpclient	🥇 29,622 (147 MiB / 37.5% CPU · 50.6 µs CPU/req)	34.5%
twp-reverse-http1	dotnet-httpclient	23,988 (117 MiB / 33.2% CPU · 55.4 µs CPU/req)	27.9%

On this macOS run origin-direct was 34.0% CPU, so the % of origin-HttpClient column is not a ceiling when that CPU is near idle. Rank those Mac peers by RPS.

Reverse peers on Block A @ 482f5984: Windows TWP is 49.8% of origin-direct and Linux TWP is 44.4%. 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	RPS	÷YARP	÷nginx
nginx-reverse-http2	dotnet-httpclient	7,670 (143 MiB / 24.8% CPU · 129.2 µs CPU/req)	0.28×	1.00×
yarp-reverse-http2	dotnet-httpclient	27,825 (105 MiB / 54.6% CPU · 78.4 µs CPU/req)	1.00×	3.63×
twp-reverse-http2-cleartext	dotnet-httpclient	🥇 30,416 (135 MiB / 50.7% CPU · 66.6 µs CPU/req)	1.09×	3.97×

Linux (ubuntu-latest)

Arm	Generator	RPS	÷YARP	÷nginx
nginx-reverse-http2	dotnet-httpclient	24,784 (100 MiB / 44.0% CPU · 71.0 µs CPU/req)	0.89×	1.00×
yarp-reverse-http2	dotnet-httpclient	27,758 (133 MiB / 50.4% CPU · 72.7 µs CPU/req)	1.00×	1.12×
twp-reverse-http2-cleartext	dotnet-httpclient	🥇 31,314 (146 MiB / 52.2% CPU · 66.7 µs CPU/req)	1.13×	1.26×

macOS (macos-15, Apple Silicon)

Arm	Generator	RPS	÷YARP	÷nginx
nginx-reverse-http2	dotnet-httpclient	🥇 33,612 (83 MiB / 25.6% CPU · 30.5 µs CPU/req)	1.08×	1.00×
yarp-reverse-http2	dotnet-httpclient	31,109 (157 MiB / 41.3% CPU · 53.1 µs CPU/req)	1.00×	0.93×
twp-reverse-http2-cleartext	dotnet-httpclient	33,241 (171 MiB / 39.8% CPU · 47.9 µs CPU/req)	1.07×	0.99×

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.

Medals are nginx / YARP / Titanium only, from the same run as Blocks A and B (37156912437). On the Linux run HAProxy was 23,420 (87 MiB / 26.7% CPU) and Envoy 3,838 (136 MiB / 24.5% CPU). On the macOS run HAProxy was 33,360 (101 MiB / 27.7% CPU) and Envoy 6,764 (129 MiB / 38.3% CPU).

Windows (windows-latest)

Arm	Generator	RPS	÷YARP	÷nginx
nginx-reverse-http3-cleartext	dotnet-httpclient	Not possible (no QUIC)	—	—
yarp-reverse-http3-cleartext	dotnet-httpclient	8,351 (154 MiB / 52.2% CPU · 249.9 µs CPU/req)	1.00×	—
twp-reverse-http3-cleartext	dotnet-httpclient	🥇 9,367 (110 MiB / 47.6% CPU · 203.1 µs CPU/req)	1.12×	—

Linux (ubuntu-latest)

Arm	Generator	RPS	÷YARP	÷nginx
nginx-reverse-http3-cleartext	dotnet-httpclient	🥇 20,832 (104 MiB / 32.3% CPU · 61.9 µs CPU/req)	1.40×	1.00×
yarp-reverse-http3-cleartext	dotnet-httpclient	14,884 (194 MiB / 50.8% CPU · 136.5 µs CPU/req)	1.00×	0.71×
twp-reverse-http3-cleartext	dotnet-httpclient	18,094 (148 MiB / 52.8% CPU · 116.7 µs CPU/req)	1.22×	0.87×

macOS (macos-15, Apple Silicon)

Arm	Generator	RPS	÷YARP	÷nginx
nginx-reverse-http3-cleartext	dotnet-httpclient	🥇 23,761 (84 MiB / 21.6% CPU · 36.4 µs CPU/req)	1.13×	1.00×
yarp-reverse-http3-cleartext	dotnet-httpclient	17,661 (213 MiB / 43.4% CPU · 98.2 µs CPU/req)	1.00×	0.88×
twp-reverse-http3-cleartext	dotnet-httpclient	15,433 (139 MiB / 43.6% CPU · 113.1 µs CPU/req)	0.87×	0.77×

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 @ 482f5984 — compare-product 37210516555. 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	nginx	YARP
HTTP/1 · plain	HTTP/1 · plain	🥇 41,064 (73 MiB / 49.7% CPU · 48.4 µs CPU/req)	25,199 (126 MiB / 24.9% CPU · 39.5 µs CPU/req)	34,970 (90 MiB / 50.1% CPU · 57.3 µs CPU/req)
HTTP/1 · plain	HTTP/1 · TLS	🥇 21,793 (88 MiB / 52.7% CPU · 96.8 µs CPU/req)	8,735 (136 MiB / 24.9% CPU · 113.9 µs CPU/req)	19,712 (100 MiB / 51.6% CPU · 104.8 µs CPU/req)
HTTP/1 · plain	HTTP/2 · plain	🥇 100,939 (113 MiB / 46.5% CPU · 18.4 µs CPU/req)	Not possible (no H2 upstream)	96,552 (103 MiB / 51.5% CPU · 21.3 µs CPU/req)
HTTP/1 · plain	HTTP/2 · TLS	🥇 32,512 (118 MiB / 45.7% CPU · 56.2 µs CPU/req)	Not possible (no H2 upstream)	30,037 (104 MiB / 47.3% CPU · 63.0 µs CPU/req)
HTTP/1 · plain	HTTP/3 · QUIC	🥇 18,652 (105 MiB / 49.8% CPU · 106.8 µs CPU/req)	Not possible (no H3 upstream)	18,139 (118 MiB / 50.6% CPU · 111.6 µs CPU/req)
HTTP/1 · TLS	HTTP/1 · plain	🥇 19,964 (84 MiB / 51.6% CPU · 103.3 µs CPU/req)	8,536 (142 MiB / 24.7% CPU · 115.7 µs CPU/req)	17,826 (101 MiB / 50.8% CPU · 114.0 µs CPU/req)
HTTP/1 · TLS	HTTP/1 · TLS	🥇 18,995 (85 MiB / 47.5% CPU · 100.0 µs CPU/req)	6,769 (144 MiB / 24.8% CPU · 146.2 µs CPU/req)	16,718 (105 MiB / 47.6% CPU · 113.9 µs CPU/req)
HTTP/1 · TLS	HTTP/2 · plain	🥇 78,921 (132 MiB / 46.6% CPU · 23.6 µs CPU/req)	Not possible (no H2 upstream)	73,677 (110 MiB / 48.6% CPU · 26.4 µs CPU/req)
HTTP/1 · TLS	HTTP/2 · TLS	🥇 25,246 (118 MiB / 43.4% CPU · 68.7 µs CPU/req)	Not possible (no H2 upstream)	24,295 (108 MiB / 45.7% CPU · 75.3 µs CPU/req)
HTTP/1 · TLS	HTTP/3 · QUIC	🥇 16,278 (109 MiB / 50.2% CPU · 123.3 µs CPU/req)	Not possible (no H3 upstream)	15,554 (124 MiB / 47.8% CPU · 122.9 µs CPU/req)
HTTP/2 · plain	HTTP/1 · plain	🥇 33,712 (120 MiB / 51.6% CPU · 61.3 µs CPU/req)	9,207 (127 MiB / 24.7% CPU · 107.4 µs CPU/req)	30,048 (92 MiB / 53.3% CPU · 70.9 µs CPU/req)
HTTP/2 · plain	HTTP/1 · TLS	🥇 29,159 (121 MiB / 52.8% CPU · 72.4 µs CPU/req)	6,709 (138 MiB / 24.8% CPU · 147.6 µs CPU/req)	26,307 (104 MiB / 53.5% CPU · 81.3 µs CPU/req)
HTTP/2 · plain	HTTP/2 · plain	🥇 101,074 (101 MiB / 34.1% CPU · 13.5 µs CPU/req)	Not possible (no H2 upstream)	73,374 (109 MiB / 54.1% CPU · 29.5 µs CPU/req)
HTTP/2 · plain	HTTP/2 · TLS	🥇 85,259 (117 MiB / 37.5% CPU · 17.6 µs CPU/req)	Not possible (no H2 upstream)	64,463 (105 MiB / 49.6% CPU · 30.7 µs CPU/req)
HTTP/2 · plain	HTTP/3 · QUIC	🥇 26,252 (150 MiB / 49.6% CPU · 75.6 µs CPU/req)	Not possible (no H3 upstream)	24,448 (124 MiB / 53.8% CPU · 88.1 µs CPU/req)
HTTP/2 · TLS	HTTP/1 · plain	🥇 30,535 (116 MiB / 52.3% CPU · 68.5 µs CPU/req)	7,738 (142 MiB / 24.8% CPU · 128.2 µs CPU/req)	27,544 (105 MiB / 54.7% CPU · 79.4 µs CPU/req)
HTTP/2 · TLS	HTTP/1 · TLS	🥇 27,044 (114 MiB / 51.6% CPU · 76.3 µs CPU/req)	6,205 (144 MiB / 24.8% CPU · 160.1 µs CPU/req)	25,109 (106 MiB / 53% CPU · 84.5 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · plain	🥇 67,582 (127 MiB / 37.1% CPU · 21.9 µs CPU/req)	Not possible (no H2 upstream)	50,109 (109 MiB / 50.1% CPU · 40.0 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · TLS	🥇 54,470 (125 MiB / 38.1% CPU · 28.0 µs CPU/req)	Not possible (no H2 upstream)	40,097 (110 MiB / 45.8% CPU · 45.7 µs CPU/req)
HTTP/2 · TLS	HTTP/3 · QUIC	🥇 24,806 (158 MiB / 50.1% CPU · 80.8 µs CPU/req)	Not possible (no H3 upstream)	22,946 (127 MiB / 50.7% CPU · 88.4 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · plain	🥇 9,045 (115 MiB / 47.6% CPU · 210.6 µs CPU/req)	Not possible (no QUIC)	7,808 (154 MiB / 50% CPU · 256.1 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 8,169 (114 MiB / 48.9% CPU · 239.2 µs CPU/req)	Not possible (no QUIC)	7,295 (162 MiB / 51.2% CPU · 281.0 µs CPU/req)
HTTP/3 · QUIC	HTTP/2 · plain	🥇 11,590 (114 MiB / 48.5% CPU · 167.5 µs CPU/req)	Not possible (no H2 upstream)	9,980 (162 MiB / 50.6% CPU · 202.9 µs CPU/req)
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 12,365 (112 MiB / 49.5% CPU · 160.2 µs CPU/req)	Not possible (no H2 upstream)	9,597 (161 MiB / 48.7% CPU · 203.0 µs CPU/req)
HTTP/3 · QUIC	HTTP/3 · QUIC	🥇 12,979 (118 MiB / 44.5% CPU · 137.1 µs CPU/req)	Not possible (no H3 upstream)	11,229 (167 MiB / 49.6% CPU · 176.8 µs CPU/req)

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 @ 482f5984 — compare-product 37210516555. 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	nginx	HAProxy	Envoy	YARP
HTTP/1 · plain	HTTP/1 · plain	31,368 (84 MiB / 49.9% CPU · 63.6 µs CPU/req)	🥇 39,240 (76 MiB / 39.6% CPU · 40.4 µs CPU/req)	33,736 (64 MiB / 43.1% CPU · 51.1 µs CPU/req)	22,017 (115 MiB / 58.4% CPU · 106.2 µs CPU/req)	28,120 (115 MiB / 49.1% CPU · 69.9 µs CPU/req)
HTTP/1 · plain	HTTP/1 · TLS	44,998 (104 MiB / 45.4% CPU · 40.4 µs CPU/req)	🥇 60,927 (93 MiB / 39.5% CPU · 25.9 µs CPU/req)	54,443 (71 MiB / 42.7% CPU · 31.4 µs CPU/req)	36,352 (118 MiB / 56% CPU · 61.6 µs CPU/req)	41,324 (127 MiB / 47.8% CPU · 46.2 µs CPU/req)
HTTP/1 · plain	HTTP/2 · plain	🥇 60,291 (135 MiB / 50.9% CPU · 33.8 µs CPU/req)	Not possible (no H2 upstream)	36,999 (65 MiB / 42.1% CPU · 45.5 µs CPU/req)	35,073 (116 MiB / 60% CPU · 68.5 µs CPU/req)	52,826 (128 MiB / 48.8% CPU · 36.9 µs CPU/req)
HTTP/1 · plain	HTTP/2 · TLS	🥇 60,446 (164 MiB / 48.4% CPU · 32.0 µs CPU/req)	Not possible (no H2 upstream)	54,142 (68 MiB / 44.3% CPU · 32.7 µs CPU/req)	42,351 (117 MiB / 57.8% CPU · 54.6 µs CPU/req)	54,565 (129 MiB / 48% CPU · 35.2 µs CPU/req)
HTTP/1 · plain	HTTP/3 · QUIC	🥇 34,726 (153 MiB / 54.3% CPU · 62.5 µs CPU/req)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	15,311 (121 MiB / 55.7% CPU · 145.4 µs CPU/req)	30,783 (162 MiB / 48.7% CPU · 63.3 µs CPU/req)
HTTP/1 · TLS	HTTP/1 · plain	40,461 (108 MiB / 48% CPU · 47.4 µs CPU/req)	🥇 51,258 (100 MiB / 39.6% CPU · 30.9 µs CPU/req)	48,775 (83 MiB / 43.7% CPU · 35.9 µs CPU/req)	32,377 (126 MiB / 57.2% CPU · 70.6 µs CPU/req)	35,042 (133 MiB / 49.7% CPU · 56.8 µs CPU/req)
HTTP/1 · TLS	HTTP/1 · TLS	19,677 (109 MiB / 48.6% CPU · 98.8 µs CPU/req)	🥇 22,947 (102 MiB / 41.6% CPU · 72.5 µs CPU/req)	21,710 (84 MiB / 43.9% CPU · 80.9 µs CPU/req)	15,563 (128 MiB / 56% CPU · 144.0 µs CPU/req)	17,121 (134 MiB / 50.2% CPU · 117.4 µs CPU/req)
HTTP/1 · TLS	HTTP/2 · plain	🥇 35,461 (137 MiB / 50.3% CPU · 56.7 µs CPU/req)	Not possible (no H2 upstream)	24,128 (82 MiB / 43.3% CPU · 71.8 µs CPU/req)	23,550 (125 MiB / 58.6% CPU · 99.5 µs CPU/req)	30,984 (139 MiB / 50% CPU · 64.6 µs CPU/req)
HTTP/1 · TLS	HTTP/2 · TLS	🥇 24,465 (151 MiB / 48% CPU · 78.5 µs CPU/req)	Not possible (no H2 upstream)	22,261 (84 MiB / 44.6% CPU · 80.2 µs CPU/req)	17,405 (125 MiB / 57.7% CPU · 132.7 µs CPU/req)	22,072 (147 MiB / 47.4% CPU · 86.0 µs CPU/req)
HTTP/1 · TLS	HTTP/3 · QUIC	🥇 18,149 (146 MiB / 52.4% CPU · 115.5 µs CPU/req)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	9,550 (130 MiB / 54.6% CPU · 228.8 µs CPU/req)	15,733 (166 MiB / 49.3% CPU · 125.2 µs CPU/req)
HTTP/2 · plain	HTTP/1 · plain	🥇 36,632 (147 MiB / 52.6% CPU · 57.4 µs CPU/req)	28,526 (78 MiB / 44.9% CPU · 63.0 µs CPU/req)	34,975 (64 MiB / 45.4% CPU · 52.0 µs CPU/req)	20,275 (116 MiB / 64.2% CPU · 126.6 µs CPU/req)	32,231 (125 MiB / 50.7% CPU · 62.9 µs CPU/req)
HTTP/2 · plain	HTTP/1 · TLS	🥇 27,860 (143 MiB / 50.8% CPU · 73.0 µs CPU/req)	22,457 (95 MiB / 45.5% CPU · 81.1 µs CPU/req)	26,956 (70 MiB / 45.1% CPU · 66.9 µs CPU/req)	17,954 (119 MiB / 60.2% CPU · 134.2 µs CPU/req)	24,772 (132 MiB / 51.7% CPU · 83.4 µs CPU/req)
HTTP/2 · plain	HTTP/2 · plain	🥇 68,356 (148 MiB / 40.2% CPU · 23.5 µs CPU/req)	Not possible (no H2 upstream)	35,625 (66 MiB / 46.5% CPU · 52.2 µs CPU/req)	25,150 (116 MiB / 62.2% CPU · 98.9 µs CPU/req)	44,979 (131 MiB / 48.9% CPU · 43.5 µs CPU/req)
HTTP/2 · plain	HTTP/2 · TLS	🥇 53,290 (149 MiB / 40.9% CPU · 30.7 µs CPU/req)	Not possible (no H2 upstream)	28,317 (67 MiB / 46.3% CPU · 65.4 µs CPU/req)	23,506 (117 MiB / 60.6% CPU · 103.2 µs CPU/req)	36,595 (132 MiB / 47.6% CPU · 52.0 µs CPU/req)
HTTP/2 · plain	HTTP/3 · QUIC	🥇 34,025 (187 MiB / 49% CPU · 57.6 µs CPU/req)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	21,067 (119 MiB / 52.3% CPU · 99.3 µs CPU/req)	31,150 (162 MiB / 47.2% CPU · 60.6 µs CPU/req)
HTTP/2 · TLS	HTTP/1 · plain	🥇 32,176 (146 MiB / 52.6% CPU · 65.4 µs CPU/req)	24,820 (100 MiB / 43.8% CPU · 70.5 µs CPU/req)	28,818 (82 MiB / 45.7% CPU · 63.4 µs CPU/req)	18,805 (128 MiB / 62.6% CPU · 133.2 µs CPU/req)	28,031 (131 MiB / 50.7% CPU · 72.3 µs CPU/req)
HTTP/2 · TLS	HTTP/1 · TLS	🥇 37,010 (153 MiB / 48.5% CPU · 52.4 µs CPU/req)	32,273 (107 MiB / 41.6% CPU · 51.5 µs CPU/req)	35,974 (83 MiB / 43.7% CPU · 48.6 µs CPU/req)	27,671 (127 MiB / 54.8% CPU · 79.3 µs CPU/req)	33,660 (135 MiB / 48.4% CPU · 57.5 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · plain	🥇 57,068 (159 MiB / 41.7% CPU · 29.2 µs CPU/req)	Not possible (no H2 upstream)	27,957 (81 MiB / 46% CPU · 65.8 µs CPU/req)	23,395 (125 MiB / 60.6% CPU · 103.6 µs CPU/req)	37,460 (133 MiB / 49% CPU · 52.3 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · TLS	🥇 47,294 (168 MiB / 41.3% CPU · 34.9 µs CPU/req)	Not possible (no H2 upstream)	26,102 (81 MiB / 45.3% CPU · 69.5 µs CPU/req)	21,911 (125 MiB / 59.8% CPU · 109.1 µs CPU/req)	31,896 (140 MiB / 47.4% CPU · 59.5 µs CPU/req)
HTTP/2 · TLS	HTTP/3 · QUIC	🥇 26,262 (193 MiB / 49.7% CPU · 75.7 µs CPU/req)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	13,400 (128 MiB / 55.5% CPU · 165.8 µs CPU/req)	23,056 (168 MiB / 47.8% CPU · 82.9 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · plain	27,227 (157 MiB / 53.2% CPU · 78.2 µs CPU/req)	30,960 (107 MiB / 31.4% CPU · 40.6 µs CPU/req)	🥇 33,254 (88 MiB / 35.8% CPU · 43.1 µs CPU/req)	13,361 (131 MiB / 54% CPU · 161.6 µs CPU/req)	21,286 (201 MiB / 50.1% CPU · 94.1 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · TLS	14,890 (161 MiB / 53.2% CPU · 142.9 µs CPU/req)	16,983 (110 MiB / 33.7% CPU · 79.5 µs CPU/req)	🥇 18,883 (89 MiB / 36.4% CPU · 77.0 µs CPU/req)	7,950 (132 MiB / 55.7% CPU · 280.2 µs CPU/req)	11,768 (208 MiB / 51.9% CPU · 176.5 µs CPU/req)
HTTP/3 · QUIC	HTTP/2 · plain	🥇 47,599 (185 MiB / 55% CPU · 46.2 µs CPU/req)	Not possible (no H2 upstream)	46,086 (88 MiB / 31.7% CPU · 27.5 µs CPU/req)	28,244 (132 MiB / 48% CPU · 67.9 µs CPU/req)	40,712 (218 MiB / 48.2% CPU · 47.4 µs CPU/req)
HTTP/3 · QUIC	HTTP/2 · TLS	19,990 (157 MiB / 52.2% CPU · 104.4 µs CPU/req)	Not possible (no H2 upstream)	🥇 20,129 (88 MiB / 36% CPU · 71.6 µs CPU/req)	9,707 (132 MiB / 58.6% CPU · 241.3 µs CPU/req)	16,272 (203 MiB / 49.2% CPU · 121.0 µs CPU/req)
HTTP/3 · QUIC	HTTP/3 · QUIC	🥇 16,696 (151 MiB / 50% CPU · 119.9 µs CPU/req)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	6,059 (132 MiB / 56% CPU · 369.7 µs CPU/req)	12,891 (205 MiB / 49.2% CPU · 152.8 µs CPU/req)

MITM (TWP only)

Same Client×Origin wires with interception on (compare-product 37210516555). 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 ≥ 0.25× and Full ≥ 0.25× reverse sustain @ c=64 (median of 3 GHA runs); reverse TWP ÷ closest peer ≥ 0.50×.

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. @ 41f4adee: 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	31,196 (91 MiB / 49.8% CPU · 63.9 µs CPU/req)	30,622 (92 MiB / 49.3% CPU · 64.4 µs CPU/req)	0.99×	0.98×
HTTP/1 · plain	HTTP/1 · TLS	44,825 (108 MiB / 46.4% CPU · 41.4 µs CPU/req)	44,084 (114 MiB / 46% CPU · 41.7 µs CPU/req)	1×	0.98×
HTTP/1 · plain	HTTP/2 · plain	58,818 (125 MiB / 52.7% CPU · 35.9 µs CPU/req)	56,959 (134 MiB / 53.3% CPU · 37.5 µs CPU/req)	0.98×	0.94×
HTTP/1 · plain	HTTP/2 · TLS	58,159 (147 MiB / 49.8% CPU · 34.2 µs CPU/req)	57,510 (157 MiB / 49.6% CPU · 34.5 µs CPU/req)	0.96×	0.95×
HTTP/1 · plain	HTTP/3 · QUIC	33,518 (148 MiB / 55.3% CPU · 66.0 µs CPU/req)	35,182 (146 MiB / 55.4% CPU · 62.9 µs CPU/req)	0.97×	1.01×
HTTP/1 · TLS	HTTP/1 · plain	39,529 (113 MiB / 48% CPU · 48.6 µs CPU/req)	38,574 (113 MiB / 47.6% CPU · 49.4 µs CPU/req)	0.98×	0.95×
HTTP/1 · TLS	HTTP/1 · TLS	19,874 (115 MiB / 49.1% CPU · 98.8 µs CPU/req)	19,276 (114 MiB / 49.1% CPU · 101.9 µs CPU/req)	1.01×	0.98×
HTTP/1 · TLS	HTTP/2 · plain	34,414 (141 MiB / 52.8% CPU · 61.3 µs CPU/req)	32,977 (141 MiB / 53% CPU · 64.2 µs CPU/req)	0.97×	0.93×
HTTP/1 · TLS	HTTP/2 · TLS	23,969 (165 MiB / 49.1% CPU · 81.9 µs CPU/req)	23,259 (160 MiB / 49.1% CPU · 84.5 µs CPU/req)	0.98×	0.95×
HTTP/1 · TLS	HTTP/3 · QUIC	18,005 (154 MiB / 52% CPU · 115.5 µs CPU/req)	16,894 (142 MiB / 53.1% CPU · 125.7 µs CPU/req)	0.99×	0.93×
HTTP/2 · plain	HTTP/1 · plain	35,085 (142 MiB / 53.8% CPU · 61.3 µs CPU/req)	34,932 (144 MiB / 53.7% CPU · 61.5 µs CPU/req)	0.96×	0.95×
HTTP/2 · plain	HTTP/1 · TLS	27,257 (147 MiB / 52% CPU · 76.3 µs CPU/req)	26,616 (153 MiB / 51.9% CPU · 78.0 µs CPU/req)	0.98×	0.96×
HTTP/2 · plain	HTTP/2 · plain	59,650 (148 MiB / 45.6% CPU · 30.6 µs CPU/req)	57,308 (149 MiB / 44.9% CPU · 31.4 µs CPU/req)	0.87×	0.84×
HTTP/2 · plain	HTTP/2 · TLS	46,667 (150 MiB / 44.6% CPU · 38.2 µs CPU/req)	45,069 (165 MiB / 44.5% CPU · 39.5 µs CPU/req)	0.88×	0.85×
HTTP/2 · plain	HTTP/3 · QUIC	34,316 (177 MiB / 50.2% CPU · 58.5 µs CPU/req)	33,710 (178 MiB / 50% CPU · 59.4 µs CPU/req)	1.01×	0.99×
HTTP/2 · TLS	HTTP/1 · plain	31,906 (147 MiB / 53.6% CPU · 67.3 µs CPU/req)	30,787 (147 MiB / 53.1% CPU · 69.0 µs CPU/req)	0.99×	0.96×
HTTP/2 · TLS	HTTP/1 · TLS	35,377 (158 MiB / 49.6% CPU · 56.0 µs CPU/req)	35,136 (151 MiB / 49.4% CPU · 56.3 µs CPU/req)	0.96×	0.95×
HTTP/2 · TLS	HTTP/2 · plain	50,844 (163 MiB / 45.9% CPU · 36.1 µs CPU/req)	48,505 (154 MiB / 45.8% CPU · 37.8 µs CPU/req)	0.89×	0.85×
HTTP/2 · TLS	HTTP/2 · TLS	41,140 (168 MiB / 44.3% CPU · 43.1 µs CPU/req)	40,144 (165 MiB / 44.4% CPU · 44.2 µs CPU/req)	0.87×	0.85×
HTTP/2 · TLS	HTTP/3 · QUIC	25,613 (199 MiB / 49.2% CPU · 76.9 µs CPU/req)	25,243 (190 MiB / 49.6% CPU · 78.5 µs CPU/req)	0.98×	0.96×
HTTP/3 · QUIC	HTTP/1 · plain	26,000 (158 MiB / 54% CPU · 83.1 µs CPU/req)	25,657 (153 MiB / 54.3% CPU · 84.7 µs CPU/req)	0.95×	0.94×
HTTP/3 · QUIC	HTTP/1 · TLS	14,382 (164 MiB / 53.4% CPU · 148.5 µs CPU/req)	13,988 (162 MiB / 53.2% CPU · 152.1 µs CPU/req)	0.97×	0.94×
HTTP/3 · QUIC	HTTP/2 · plain	45,075 (185 MiB / 55.8% CPU · 49.5 µs CPU/req)	45,689 (188 MiB / 55.1% CPU · 48.2 µs CPU/req)	0.95×	0.96×
HTTP/3 · QUIC	HTTP/2 · TLS	19,839 (155 MiB / 53.3% CPU · 107.5 µs CPU/req)	18,948 (160 MiB / 52.6% CPU · 111.1 µs CPU/req)	0.99×	0.95×
HTTP/3 · QUIC	HTTP/3 · QUIC	16,001 (153 MiB / 52.7% CPU · 131.8 µs CPU/req)	15,955 (156 MiB / 52.9% CPU · 132.6 µs CPU/req)	0.96×	0.96×

macOS — Titanium vs nginx vs HAProxy vs Envoy vs YARP

Reverse

Median of 3 repeats on macos-15 (Apple Silicon M1, 3-core / 7 GB). Bare reverse 5×5 @ 482f5984 — compare-product 37210516555. 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-arm64 1.36.7), Homebrew libmsquic (+ DYLD_*), and YARP. Use the pinned macos-15 label, not macos-latest. 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	nginx	HAProxy	Envoy	YARP
HTTP/1 · plain	HTTP/1 · plain	7,105 (114 MiB / 10.7% CPU · 45.3 µs CPU/req)	🥇 17,906 (68 MiB / 11.6% CPU · 19.4 µs CPU/req)	7,751 (91 MiB / 10.3% CPU · 40.0 µs CPU/req)	7,226 (111 MiB / 24.3% CPU · 101.0 µs CPU/req)	9,245 (169 MiB / 13.6% CPU · 44.1 µs CPU/req)
HTTP/1 · plain	HTTP/1 · TLS	6,570 (126 MiB / 15.5% CPU · 70.6 µs CPU/req)	12,385 (78 MiB / 16.3% CPU · 39.5 µs CPU/req)	🥇 13,060 (96 MiB / 21.3% CPU · 49.0 µs CPU/req)	9,394 (113 MiB / 29% CPU · 92.6 µs CPU/req)	5,214 (203 MiB / 13.6% CPU · 78.4 µs CPU/req)
HTTP/1 · plain	HTTP/2 · plain	🥇 45,183 (144 MiB / 26.9% CPU · 17.9 µs CPU/req)	Not possible (no H2 upstream)	34,431 (104 MiB / 30.4% CPU · 26.5 µs CPU/req)	34,198 (111 MiB / 34.5% CPU · 30.3 µs CPU/req)	36,843 (163 MiB / 21.4% CPU · 17.4 µs CPU/req)
HTTP/1 · plain	HTTP/2 · TLS	17,960 (164 MiB / 17.5% CPU · 29.2 µs CPU/req)	Not possible (no H2 upstream)	13,000 (99 MiB / 13.4% CPU · 30.8 µs CPU/req)	10,582 (113 MiB / 19.7% CPU · 55.9 µs CPU/req)	🥇 18,772 (167 MiB / 16.3% CPU · 26.1 µs CPU/req)
HTTP/1 · plain	HTTP/3 · QUIC	11,126 (133 MiB / 23.9% CPU · 64.5 µs CPU/req)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	3,729 (117 MiB / 18.5% CPU · 148.9 µs CPU/req)	🥇 11,631 (163 MiB / 19% CPU · 49.1 µs CPU/req)
HTTP/1 · TLS	HTTP/1 · plain	6,420 (128 MiB / 11.6% CPU · 54.2 µs CPU/req)	🥇 18,160 (85 MiB / 15.8% CPU · 26.0 µs CPU/req)	15,854 (108 MiB / 19.8% CPU · 37.4 µs CPU/req)	8,061 (122 MiB / 20.2% CPU · 75.1 µs CPU/req)	10,810 (162 MiB / 20.4% CPU · 56.6 µs CPU/req)
HTTP/1 · TLS	HTTP/1 · TLS	🥇 7,991 (153 MiB / 20.5% CPU · 76.8 µs CPU/req)	7,624 (89 MiB / 11.2% CPU · 43.9 µs CPU/req)	6,204 (109 MiB / 13.3% CPU · 64.2 µs CPU/req)	6,747 (123 MiB / 25% CPU · 111.1 µs CPU/req)	3,900 (224 MiB / 11.9% CPU · 91.3 µs CPU/req)
HTTP/1 · TLS	HTTP/2 · plain	11,822 (143 MiB / 11% CPU · 27.8 µs CPU/req)	Not possible (no H2 upstream)	9,010 (112 MiB / 14.4% CPU · 47.9 µs CPU/req)	22,406 (120 MiB / 27.8% CPU · 37.2 µs CPU/req)	🥇 27,679 (172 MiB / 22.6% CPU · 24.5 µs CPU/req)
HTTP/1 · TLS	HTTP/2 · TLS	28,926 (197 MiB / 35.6% CPU · 36.9 µs CPU/req)	Not possible (no H2 upstream)	23,046 (131 MiB / 37.1% CPU · 48.3 µs CPU/req)	24,438 (121 MiB / 48% CPU · 58.9 µs CPU/req)	🥇 34,479 (171 MiB / 36.4% CPU · 31.7 µs CPU/req)
HTTP/1 · TLS	HTTP/3 · QUIC	🥇 5,392 (161 MiB / 13.9% CPU · 77.2 µs CPU/req)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	4,701 (125 MiB / 20.7% CPU · 132.3 µs CPU/req)	5,267 (182 MiB / 10.7% CPU · 60.9 µs CPU/req)
HTTP/2 · plain	HTTP/1 · plain	22,542 (173 MiB / 22.7% CPU · 30.2 µs CPU/req)	18,571 (70 MiB / 14.2% CPU · 23.0 µs CPU/req)	🥇 28,814 (98 MiB / 26.2% CPU · 27.3 µs CPU/req)	15,206 (112 MiB / 34.8% CPU · 68.6 µs CPU/req)	18,839 (153 MiB / 24.5% CPU · 39.0 µs CPU/req)
HTTP/2 · plain	HTTP/1 · TLS	🥇 15,111 (200 MiB / 20.4% CPU · 40.5 µs CPU/req)	7,653 (81 MiB / 7.8% CPU · 30.7 µs CPU/req)	8,962 (97 MiB / 12.7% CPU · 42.6 µs CPU/req)	11,431 (114 MiB / 26% CPU · 68.2 µs CPU/req)	9,622 (164 MiB / 22.4% CPU · 69.7 µs CPU/req)
HTTP/2 · plain	HTTP/2 · plain	19,941 (220 MiB / 9.9% CPU · 14.9 µs CPU/req)	Not possible (no H2 upstream)	11,895 (101 MiB / 12.7% CPU · 32.0 µs CPU/req)	21,864 (110 MiB / 24.9% CPU · 34.1 µs CPU/req)	🥇 23,982 (163 MiB / 17.6% CPU · 22.0 µs CPU/req)
HTTP/2 · plain	HTTP/2 · TLS	🥇 31,236 (240 MiB / 14.2% CPU · 13.6 µs CPU/req)	Not possible (no H2 upstream)	25,521 (104 MiB / 24.2% CPU · 28.4 µs CPU/req)	17,483 (111 MiB / 18% CPU · 30.8 µs CPU/req)	17,015 (171 MiB / 10.6% CPU · 18.7 µs CPU/req)
HTTP/2 · plain	HTTP/3 · QUIC	7,398 (206 MiB / 13.8% CPU · 55.9 µs CPU/req)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	12,701 (116 MiB / 31.4% CPU · 74.2 µs CPU/req)	🥇 20,539 (167 MiB / 26.4% CPU · 38.6 µs CPU/req)
HTTP/2 · TLS	HTTP/1 · plain	19,283 (181 MiB / 23.9% CPU · 37.2 µs CPU/req)	19,033 (83 MiB / 22.6% CPU · 35.7 µs CPU/req)	🥇 25,232 (118 MiB / 33.6% CPU · 39.9 µs CPU/req)	17,737 (121 MiB / 46.6% CPU · 78.9 µs CPU/req)	12,200 (159 MiB / 21.8% CPU · 53.5 µs CPU/req)
HTTP/2 · TLS	HTTP/1 · TLS	6,993 (216 MiB / 14.6% CPU · 62.6 µs CPU/req)	7,862 (88 MiB / 9.2% CPU · 35.3 µs CPU/req)	🥇 15,084 (110 MiB / 24.3% CPU · 48.4 µs CPU/req)	8,194 (122 MiB / 19.7% CPU · 72.0 µs CPU/req)	11,945 (166 MiB / 20.8% CPU · 52.1 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · plain	9,534 (238 MiB / 7.7% CPU · 24.4 µs CPU/req)	Not possible (no H2 upstream)	12,695 (111 MiB / 20.9% CPU · 49.5 µs CPU/req)	7,376 (119 MiB / 16.5% CPU · 67.2 µs CPU/req)	🥇 16,833 (248 MiB / 13.4% CPU · 23.8 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · TLS	12,624 (265 MiB / 8.1% CPU · 19.2 µs CPU/req)	Not possible (no H2 upstream)	🥇 29,595 (121 MiB / 27.6% CPU · 27.9 µs CPU/req)	11,210 (120 MiB / 14.8% CPU · 39.5 µs CPU/req)	6,121 (184 MiB / 6.8% CPU · 33.4 µs CPU/req)
HTTP/2 · TLS	HTTP/3 · QUIC	5,549 (208 MiB / 13.5% CPU · 72.9 µs CPU/req)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	7,792 (123 MiB / 23.3% CPU · 89.8 µs CPU/req)	🥇 10,099 (172 MiB / 17.1% CPU · 50.9 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · plain	6,298 (145 MiB / 23.7% CPU · 112.8 µs CPU/req)	🥇 12,690 (86 MiB / 12.2% CPU · 28.8 µs CPU/req)	10,825 (104 MiB / 15% CPU · 41.5 µs CPU/req)	4,599 (126 MiB / 24.3% CPU · 158.5 µs CPU/req)	3,324 (252 MiB / 14.2% CPU · 127.9 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 10,902 (162 MiB / 26.8% CPU · 73.9 µs CPU/req)	Not measured	9,558 (108 MiB / 19.9% CPU · 62.4 µs CPU/req)	4,522 (126 MiB / 28% CPU · 185.9 µs CPU/req)	3,907 (267 MiB / 20.6% CPU · 158.5 µs CPU/req)
HTTP/3 · QUIC	HTTP/2 · plain	5,108 (132 MiB / 16.1% CPU · 94.6 µs CPU/req)	Not possible (no H2 upstream)	5,022 (108 MiB / 11% CPU · 65.9 µs CPU/req)	4,249 (124 MiB / 25% CPU · 176.8 µs CPU/req)	🥇 5,792 (266 MiB / 14.8% CPU · 76.9 µs CPU/req)
HTTP/3 · QUIC	HTTP/2 · TLS	7,887 (142 MiB / 17.7% CPU · 67.2 µs CPU/req)	Not possible (no H2 upstream)	7,318 (107 MiB / 14.8% CPU · 60.7 µs CPU/req)	5,970 (122 MiB / 28.2% CPU · 141.7 µs CPU/req)	🥇 8,025 (275 MiB / 17.7% CPU · 66.3 µs CPU/req)
HTTP/3 · QUIC	HTTP/3 · QUIC	4,485 (137 MiB / 17% CPU · 113.9 µs CPU/req)	Not possible (no H3 upstream)	Not possible (no H3 upstream)	🥇 6,850 (122 MiB / 34.9% CPU · 152.7 µs CPU/req)	3,931 (253 MiB / 20.4% CPU · 156.0 µs CPU/req)

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/cache) runs on H1 terminate-lite without SessionEventArgs; a cache hit skips the origin. JWT caches successful bearer validations. CORS only adds response headers on the way out, so it stays on terminate-lite. Circuit breaker and idempotent retry have to see the request or the status code, so they stay on the session path and should land near the intercept row. Maintainer gate thresholds live under Maintainer notes.

Median of 3 repeats on ubuntu-latest. Every row in this table was re-measured @ 482f5984 (Actions 37210543774). Warmup 2s / measure 8s; concurrency 8–64; sustain = median peak RPS among SLO-pass steps @ c=64. RPS cells show sustain; <sub> holds peak (when higher) plus (MiB / CPU%). The Gate column is that script's floor.

Arm	Linux	Lin÷	Gate
twp-cli-reverse-http1 vs library	33,245 (164 MiB / 49.9% CPU · 60.0 µs CPU/req)	1.05×	≥ 0.50×
twp-cli-reverse-http1-tls vs library TLS	24,014 (195 MiB / 50.4% CPU · 84.0 µs CPU/req)	1.03×	≥ 0.50×
twp-cli-reverse-http1-route vs CLI	63,157 (169 MiB / 49.0% CPU · 31.0 µs CPU/req)	0.96×	≥ 0.50×
twp-cli-plus-base-http1 vs CLI	48,080 (170 MiB / 49.4% CPU · 41.1 µs CPU/req)	0.99×	≥ 0.50×
twp-cli-plus-cache-http1 (cold) vs CLI	58,197 (168 MiB / 47.7% CPU · 32.8 µs CPU/req)	1.74×	≥ 0.50×
twp-cli-intercept-http1 vs CLI	24,191 (176 MiB / 53.0% CPU · 87.6 µs CPU/req)	0.73×	≥ 0.50×
twp-cli-plus-waf-http1 vs CLI	32,992 (169 MiB / 50.4% CPU · 61.1 µs CPU/req)	0.97×	≥ 0.50×
twp-cli-plus-cidr-http1 vs CLI	64,841 (235 MiB / 48.2% CPU · 29.8 µs CPU/req)	1.01×	≥ 0.50×
twp-cli-plus-jwt-http1 vs CLI	30,337 (192 MiB / 50.4% CPU · 66.4 µs CPU/req)	0.91×	≥ 0.50×
twp-cli-plus-ratelimit-http1 vs CLI	62,173 (235 MiB / 48.0% CPU · 30.9 µs CPU/req)	0.97×	≥ 0.50×
twp-cli-plus-resilience-http1 vs CLI	55,849 (240 MiB / 48.8% CPU · 34.9 µs CPU/req)	1.03×	≥ 0.50×
twp-cli-plus-discovery-file-http1 vs CLI	37,627 (169 MiB / 49.4% CPU · 52.5 µs CPU/req)	1.01×	≥ 0.50×
twp-cli-plus-metrics-scrape-http1 vs CLI	33,316 (174 MiB / 49.7% CPU · 59.7 µs CPU/req)	1.01×	≥ 0.50×
twp-cli-plus-cache-hit-http1 vs cache cold	59,678 (171 MiB / 49.1% CPU · 32.9 µs CPU/req)	1.03×	≥ 0.50×
twp-cli-plus-cors-http1 vs CLI	31,610 (166 MiB / 50.9% CPU · 64.4 µs CPU/req)	0.98×	≥ 0.50×
twp-cli-plus-circuit-http1 vs CLI	51,728 (175 MiB / 49.6% CPU · 38.4 µs CPU/req)	0.82×	≥ 0.50×
twp-cli-plus-retry-http1 vs CLI	26,733 (174 MiB / 50.5% CPU · 75.5 µs CPU/req)	0.82×	≥ 0.50×
twp-cli-static-http1 vs CLI	111,716 (166 MiB / 48.1% CPU · 17.2 µs CPU/req)	1.74×	≥ 0.50×
twp-cli-logging-http1 vs CLI	33,316 (163 MiB / 50.2% CPU · 60.3 µs CPU/req)	1.00×	≥ 0.50×
twp-cli-lb-leasttime-http1 vs route	31,121 (193 MiB / 51.8% CPU · 66.6 µs CPU/req)	0.94×	≥ 0.50×
twp-cli-dialect-twp-http1 vs CLI	33,668 (161 MiB / 49.7% CPU · 59.0 µs CPU/req)	1.02×	≥ 0.50×

validate-edition-gates.ps1 floors are 0.50×. Each ÷ column uses the two arms from the same job. Circuit breaker and idempotent retry stay on the session path, so a ratio near intercept is expected. Laptop smoke ratios stay on Performance Local Lab — Editions.

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). 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 — ratios vs YARP are in the tables below (@ 482f5984), unlike the tiny-GET H2↔H2 medals above. Those cells also show goodput (MiB/s) and µs CPU/req, and the medal on those rows follows goodput.

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 is loss% only (no per-datagram delay) + drops (QUIC / MsQuic-safe). 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 @ 482f5984. Source: Actions 37210521790 (compare-bodies). Warmup 2s / measure 8s. RPS cells include (MiB / CPU%) footprints.

Not possible: HAProxy and Envoy columns are omitted (no official Windows port).

Body	Client	Origin	TWP	nginx	YARP
64 KiB	HTTP/1 · TLS	HTTP/1 · plain	🥇 9,157 (89 MiB / 46.1% CPU · 572 MiB/s · 201.2 µs CPU/req)	619 (143 MiB / 24.8% CPU · 39 MiB/s · 1599.1 µs CPU/req)	7,890 (134 MiB / 46.9% CPU · 493 MiB/s · 237.7 µs CPU/req)
64 KiB	HTTP/2 · TLS	HTTP/1 · plain	🥇 8,611 (217 MiB / 51.0% CPU · 538 MiB/s · 237.1 µs CPU/req)	584 (142 MiB / 24.8% CPU · 36 MiB/s · 1698.4 µs CPU/req)	8,259 (126 MiB / 51.6% CPU · 516 MiB/s · 250.0 µs CPU/req)
64 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 4,524 (139 MiB / 45.4% CPU · 283 MiB/s · 401.3 µs CPU/req)	Not possible (no QUIC)	2,379 (183 MiB / 51.3% CPU · 149 MiB/s · 862.8 µs CPU/req)
256 KiB	HTTP/1 · TLS	HTTP/1 · plain	🥇 2,857 (120 MiB / 46.5% CPU · 714 MiB/s · 650.8 µs CPU/req)	166 (142 MiB / 24.8% CPU · 41 MiB/s · 5987.9 µs CPU/req)	2,616 (136 MiB / 48.7% CPU · 654 MiB/s · 744.5 µs CPU/req)
256 KiB	HTTP/2 · TLS	HTTP/1 · plain	3,083 (203 MiB / 48.3% CPU · 771 MiB/s · 626.7 µs CPU/req)	220 (143 MiB / 24.9% CPU · 55 MiB/s · 4517.2 µs CPU/req)	🥇 3,226 (124 MiB / 46.7% CPU · 807 MiB/s · 579.0 µs CPU/req)
256 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 1,995 (118 MiB / 42.1% CPU · 499 MiB/s · 843.9 µs CPU/req)	Not possible (no QUIC)	1,469 (182 MiB / 42.0% CPU · 367 MiB/s · 1143.3 µs CPU/req)
64 KiB	HTTP/2 · plain	HTTP/1 · plain	🥇 12,018 (191 MiB / 44.0% CPU · 751 MiB/s · 146.5 µs CPU/req)	1,999 (127 MiB / 24.7% CPU · 125 MiB/s · 493.9 µs CPU/req)	10,459 (121 MiB / 50.7% CPU · 654 MiB/s · 194.0 µs CPU/req)
64 KiB	HTTP/2 · TLS	HTTP/2 · plain	🥇 8,329 (169 MiB / 45.9% CPU · 521 MiB/s · 220.6 µs CPU/req)	Not possible	7,536 (131 MiB / 52.7% CPU · 471 MiB/s · 279.5 µs CPU/req)
64 KiB	HTTP/2 · TLS	HTTP/2 · TLS	🥇 6,344 (168 MiB / 41.4% CPU · 396 MiB/s · 261.1 µs CPU/req)	Not possible	5,415 (132 MiB / 49.2% CPU · 338 MiB/s · 363.1 µs CPU/req)
64 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	🥇 4,650 (135 MiB / 43.7% CPU · 291 MiB/s · 376.0 µs CPU/req)	Not possible	2,836 (186 MiB / 52.7% CPU · 177 MiB/s · 743.1 µs CPU/req)
64 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	🥇 3,290 (146 MiB / 46.4% CPU · 206 MiB/s · 564.1 µs CPU/req)	Not possible (no QUIC)	1,821 (196 MiB / 50.2% CPU · 114 MiB/s · 1101.7 µs CPU/req)
256 KiB	HTTP/2 · plain	HTTP/1 · plain	🥇 4,300 (176 MiB / 34.6% CPU · 1,075 MiB/s · 322.1 µs CPU/req)	654 (127 MiB / 24.9% CPU · 163 MiB/s · 1522.6 µs CPU/req)	3,909 (116 MiB / 42.2% CPU · 977 MiB/s · 432.3 µs CPU/req)
256 KiB	HTTP/2 · TLS	HTTP/2 · plain	3,668 (203 MiB / 44.6% CPU · 917 MiB/s · 486.0 µs CPU/req)	Not possible	🥇 3,797 (170 MiB / 44.1% CPU · 949 MiB/s · 464.1 µs CPU/req)
256 KiB	HTTP/2 · TLS	HTTP/2 · TLS	🥇 2,082 (200 MiB / 42.7% CPU · 521 MiB/s · 819.4 µs CPU/req)	Not possible	1,768 (146 MiB / 42.8% CPU · 442 MiB/s · 968.1 µs CPU/req)
256 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	870 (145 MiB / 49.1% CPU · 217 MiB/s · 2258.0 µs CPU/req)	Not possible	🥇 1,338 (194 MiB / 48.6% CPU · 334 MiB/s · 1452.8 µs CPU/req)
256 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	771 (peak 1,607 · 122 MiB / 43.9% CPU · 193 MiB/s · 2280.8 µs CPU/req)	Not possible (no QUIC)	🥇 1,514 (186 MiB / 46.1% CPU · 379 MiB/s · 1216.8 µs CPU/req)

nginx/Windows collapses on large reverse bodies in this harness; treat as same-OS only. Prefer the ratio columns; absolute RPS swings by VM.

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

Median of 3 repeats @ 482f5984. Source: Actions 37210521790 (compare-bodies). Warmup 2s / measure 8s.

Body	Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
64 KiB	HTTP/1 · TLS	HTTP/1 · plain	🥇 12,766 (127 MiB / 41.8% CPU · 798 MiB/s · 130.9 µs CPU/req)	8,147 (100 MiB / 49.4% CPU · 509 MiB/s · 242.6 µs CPU/req)	12,654 (88 MiB / 39.8% CPU · 791 MiB/s · 125.9 µs CPU/req)	11,049 (131 MiB / 43.7% CPU · 691 MiB/s · 158.2 µs CPU/req)	9,584 (166 MiB / 46.8% CPU · 599 MiB/s · 195.2 µs CPU/req)
64 KiB	HTTP/2 · TLS	HTTP/1 · plain	14,876 (245 MiB / 42.9% CPU · 930 MiB/s · 115.3 µs CPU/req)	9,240 (102 MiB / 48.3% CPU · 578 MiB/s · 209.2 µs CPU/req)	12,108 (87 MiB / 45.4% CPU · 757 MiB/s · 150.1 µs CPU/req)	🥇 15,828 (134 MiB / 44.0% CPU · 989 MiB/s · 111.2 µs CPU/req)	12,621 (157 MiB / 45.8% CPU · 789 MiB/s · 145.1 µs CPU/req)
64 KiB	HTTP/3 · QUIC	HTTP/1 · plain	5,606 (203 MiB / 45.3% CPU · 350 MiB/s · 323.3 µs CPU/req)	2,589 (116 MiB / 28.7% CPU · 162 MiB/s · 443.7 µs CPU/req)	🥇 5,814 (94 MiB / 36.3% CPU · 363 MiB/s · 250.0 µs CPU/req)	4,457 (138 MiB / 51.2% CPU · 279 MiB/s · 459.6 µs CPU/req)	3,738 (237 MiB / 51.5% CPU · 234 MiB/s · 551.6 µs CPU/req)
256 KiB	HTTP/1 · TLS	HTTP/1 · plain	3,855 (123 MiB / 31.7% CPU · 964 MiB/s · 328.6 µs CPU/req)	2,448 (100 MiB / 48.4% CPU · 612 MiB/s · 790.1 µs CPU/req)	🥇 4,131 (87 MiB / 27.6% CPU · 1,033 MiB/s · 267.7 µs CPU/req)	3,563 (145 MiB / 31.7% CPU · 891 MiB/s · 356.3 µs CPU/req)	3,031 (166 MiB / 41.9% CPU · 758 MiB/s · 552.4 µs CPU/req)
256 KiB	HTTP/2 · TLS	HTTP/1 · plain	2,052 (210 MiB / 41.7% CPU · 513 MiB/s · 812.5 µs CPU/req)	1,546 (99 MiB / 50.5% CPU · 386 MiB/s · 1308.1 µs CPU/req)	2,086 (87 MiB / 41.5% CPU · 522 MiB/s · 795.0 µs CPU/req)	🥇 2,412 (152 MiB / 36.1% CPU · 603 MiB/s · 599.1 µs CPU/req)	1,813 (164 MiB / 46.1% CPU · 453 MiB/s · 1016.8 µs CPU/req)
256 KiB	HTTP/3 · QUIC	HTTP/1 · plain	1,600 (174 MiB / 49.6% CPU · 400 MiB/s · 1241.4 µs CPU/req)	746 (116 MiB / 28.5% CPU · 187 MiB/s · 1526.1 µs CPU/req)	🥇 1,932 (95 MiB / 34.8% CPU · 483 MiB/s · 720.7 µs CPU/req)	1,634 (154 MiB / 45.3% CPU · 408 MiB/s · 1109.7 µs CPU/req)	1,284 (238 MiB / 51.7% CPU · 321 MiB/s · 1611.7 µs CPU/req)
64 KiB	HTTP/2 · plain	HTTP/1 · plain	🥇 11,786 (262 MiB / 43.3% CPU · 737 MiB/s · 147.0 µs CPU/req)	6,946 (78 MiB / 50.4% CPU · 434 MiB/s · 290.5 µs CPU/req)	10,740 (72 MiB / 44.3% CPU · 671 MiB/s · 165.1 µs CPU/req)	10,789 (124 MiB / 47.5% CPU · 674 MiB/s · 176.0 µs CPU/req)	10,911 (150 MiB / 45.0% CPU · 682 MiB/s · 164.9 µs CPU/req)
64 KiB	HTTP/2 · TLS	HTTP/2 · plain	🥇 8,816 (208 MiB / 36.5% CPU · 551 MiB/s · 165.4 µs CPU/req)	Not possible	7,247 (91 MiB / 42.0% CPU · 453 MiB/s · 231.9 µs CPU/req)	8,232 (136 MiB / 43.0% CPU · 515 MiB/s · 209.1 µs CPU/req)	6,821 (195 MiB / 48.0% CPU · 426 MiB/s · 281.4 µs CPU/req)
64 KiB	HTTP/2 · TLS	HTTP/2 · TLS	🥇 5,010 (206 MiB / 41.6% CPU · 313 MiB/s · 331.8 µs CPU/req)	Not possible	4,209 (91 MiB / 44.7% CPU · 263 MiB/s · 424.4 µs CPU/req)	4,765 (141 MiB / 43.9% CPU · 298 MiB/s · 368.1 µs CPU/req)	3,614 (180 MiB / 46.7% CPU · 226 MiB/s · 517.1 µs CPU/req)
64 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	3,400 (181 MiB / 45.5% CPU · 213 MiB/s · 535.0 µs CPU/req)	Not possible	🥇 3,644 (99 MiB / 40.4% CPU · 228 MiB/s · 443.2 µs CPU/req)	2,851 (145 MiB / 49.2% CPU · 178 MiB/s · 690.5 µs CPU/req)	2,478 (248 MiB / 49.2% CPU · 155 MiB/s · 794.2 µs CPU/req)
64 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	3,756 (214 MiB / 45.7% CPU · 235 MiB/s · 486.5 µs CPU/req)	2,286 (117 MiB / 38.5% CPU · 143 MiB/s · 673.3 µs CPU/req)	🥇 3,924 (95 MiB / 41.1% CPU · 245 MiB/s · 419.5 µs CPU/req)	3,254 (138 MiB / 49.6% CPU · 203 MiB/s · 609.5 µs CPU/req)	2,762 (248 MiB / 51.5% CPU · 173 MiB/s · 745.3 µs CPU/req)
256 KiB	HTTP/2 · plain	HTTP/1 · plain	3,834 (205 MiB / 34.4% CPU · 959 MiB/s · 359.0 µs CPU/req)	2,872 (78 MiB / 47.4% CPU · 718 MiB/s · 660.1 µs CPU/req)	4,154 (72 MiB / 30.5% CPU · 1,038 MiB/s · 293.3 µs CPU/req)	🥇 4,238 (143 MiB / 30.0% CPU · 1,060 MiB/s · 283.0 µs CPU/req)	3,745 (151 MiB / 36.0% CPU · 936 MiB/s · 384.0 µs CPU/req)
256 KiB	HTTP/2 · TLS	HTTP/2 · plain	🥇 2,137 (227 MiB / 37.9% CPU · 534 MiB/s · 708.9 µs CPU/req)	Not possible	1,978 (99 MiB / 39.2% CPU · 494 MiB/s · 792.4 µs CPU/req)	2,052 (146 MiB / 37.5% CPU · 513 MiB/s · 731.7 µs CPU/req)	1,553 (206 MiB / 45.2% CPU · 388 MiB/s · 1165.8 µs CPU/req)
256 KiB	HTTP/2 · TLS	HTTP/2 · TLS	🥇 1,368 (272 MiB / 40.3% CPU · 342 MiB/s · 1179.3 µs CPU/req)	Not possible	1,282 (97 MiB / 40.1% CPU · 321 MiB/s · 1249.4 µs CPU/req)	1,351 (143 MiB / 37.9% CPU · 338 MiB/s · 1121.2 µs CPU/req)	963 (181 MiB / 44.2% CPU · 241 MiB/s · 1836.8 µs CPU/req)
256 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	1,125 (204 MiB / 55.9% CPU · 281 MiB/s · 1988.4 µs CPU/req)	Not possible	🥇 2,020 (105 MiB / 33.9% CPU · 505 MiB/s · 671.7 µs CPU/req)	1,706 (156 MiB / 41.3% CPU · 426 MiB/s · 968.0 µs CPU/req)	1,328 (238 MiB / 47.5% CPU · 332 MiB/s · 1431.3 µs CPU/req)
256 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	985 (192 MiB / 49.7% CPU · 246 MiB/s · 2018.7 µs CPU/req)	605 (117 MiB / 38.6% CPU · 151 MiB/s · 2550.8 µs CPU/req)	🥇 1,140 (98 MiB / 39.8% CPU · 285 MiB/s · 1396.2 µs CPU/req)	1,081 (153 MiB / 44.0% CPU · 270 MiB/s · 1628.9 µs CPU/req)	796 (256 MiB / 49.9% CPU · 199 MiB/s · 2507.3 µs CPU/req)

Prefer the ratio columns; absolute RPS swings by VM.

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

Median of 3 repeats on macos-15 (Apple Silicon M1, 3-core / 7 GB) @ 482f5984. Source: Actions 37210521790 (compare-bodies). Warmup 2s / measure 8s.

Body	Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
64 KiB	HTTP/1 · TLS	HTTP/1 · plain	4,134 (177 MiB / 14.9% CPU · 258 MiB/s · 108.4 µs CPU/req)	🥇 9,185 (84 MiB / 21.7% CPU · 574 MiB/s · 70.9 µs CPU/req)	9,097 (120 MiB / 25.8% CPU · 569 MiB/s · 85.0 µs CPU/req)	7,800 (133 MiB / 26.8% CPU · 488 MiB/s · 103.1 µs CPU/req)	5,900 (183 MiB / 27.1% CPU · 369 MiB/s · 138.1 µs CPU/req)
64 KiB	HTTP/2 · TLS	HTTP/1 · plain	3,149 (469 MiB / 26.1% CPU · 197 MiB/s · 249.0 µs CPU/req)	4,113 (87 MiB / 13.3% CPU · 257 MiB/s · 97.2 µs CPU/req)	4,297 (120 MiB / 23.9% CPU · 269 MiB/s · 167.2 µs CPU/req)	🥇 5,386 (135 MiB / 20.3% CPU · 337 MiB/s · 113.0 µs CPU/req)	2,376 (171 MiB / 16.5% CPU · 149 MiB/s · 208.4 µs CPU/req)
64 KiB	HTTP/3 · QUIC	HTTP/1 · plain	2,786 (278 MiB / 29.7% CPU · 174 MiB/s · 319.8 µs CPU/req)	🥇 3,796 (97 MiB / 28.1% CPU · 237 MiB/s · 221.9 µs CPU/req)	2,692 (129 MiB / 31.3% CPU · 168 MiB/s · 348.8 µs CPU/req)	2,141 (132 MiB / 44.8% CPU · 134 MiB/s · 627.5 µs CPU/req)	2,136 (446 MiB / 37.1% CPU · 133 MiB/s · 521.2 µs CPU/req)
256 KiB	HTTP/1 · TLS	HTTP/1 · plain	2,115 (215 MiB / 25.1% CPU · 529 MiB/s · 355.5 µs CPU/req)	1,844 (84 MiB / 18.5% CPU · 461 MiB/s · 301.6 µs CPU/req)	🥇 2,503 (117 MiB / 20.9% CPU · 626 MiB/s · 250.0 µs CPU/req)	2,018 (154 MiB / 21.3% CPU · 504 MiB/s · 316.9 µs CPU/req)	1,527 (183 MiB / 27.6% CPU · 382 MiB/s · 541.4 µs CPU/req)
256 KiB	HTTP/2 · TLS	HTTP/1 · plain	1,053 (447 MiB / 19.2% CPU · 263 MiB/s · 548.5 µs CPU/req)	1,725 (85 MiB / 22.6% CPU · 431 MiB/s · 393.0 µs CPU/req)	2,034 (117 MiB / 28.1% CPU · 509 MiB/s · 413.9 µs CPU/req)	🥇 2,175 (154 MiB / 27.0% CPU · 544 MiB/s · 372.7 µs CPU/req)	1,353 (178 MiB / 31.0% CPU · 338 MiB/s · 687.2 µs CPU/req)
256 KiB	HTTP/3 · QUIC	HTTP/1 · plain	508 (190 MiB / 32.1% CPU · 127 MiB/s · 1893.7 µs CPU/req)	🥇 841 (95 MiB / 28.5% CPU · 210 MiB/s · 1017.0 µs CPU/req)	515 (peak 521 · 106 MiB / 29.0% CPU · 129 MiB/s · 1691.0 µs CPU/req)	530 (159 MiB / 51.6% CPU · 132 MiB/s · 2920.7 µs CPU/req)	256 (211 MiB / 29.4% CPU · 64 MiB/s · 3446.0 µs CPU/req)
64 KiB	HTTP/2 · plain	HTTP/1 · plain	🥇 9,299 (455 MiB / 22.4% CPU · 581 MiB/s · 72.4 µs CPU/req)	7,686 (73 MiB / 12.3% CPU · 480 MiB/s · 47.9 µs CPU/req)	8,105 (105 MiB / 20.6% CPU · 507 MiB/s · 76.4 µs CPU/req)	7,881 (135 MiB / 21.8% CPU · 493 MiB/s · 82.8 µs CPU/req)	7,453 (187 MiB / 21.8% CPU · 466 MiB/s · 87.7 µs CPU/req)
64 KiB	HTTP/2 · TLS	HTTP/2 · plain	🥇 10,070 (394 MiB / 36.2% CPU · 629 MiB/s · 107.9 µs CPU/req)	Not possible	6,370 (127 MiB / 34.9% CPU · 398 MiB/s · 164.5 µs CPU/req)	6,797 (126 MiB / 34.7% CPU · 425 MiB/s · 153.1 µs CPU/req)	7,814 (272 MiB / 47.4% CPU · 488 MiB/s · 181.9 µs CPU/req)
64 KiB	HTTP/2 · TLS	HTTP/2 · TLS	2,166 (386 MiB / 15.5% CPU · 135 MiB/s · 214.1 µs CPU/req)	Not possible	🥇 3,441 (123 MiB / 26.2% CPU · 215 MiB/s · 228.1 µs CPU/req)	3,436 (126 MiB / 24.7% CPU · 215 MiB/s · 215.7 µs CPU/req)	2,101 (253 MiB / 20.7% CPU · 131 MiB/s · 296.1 µs CPU/req)
64 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	🥇 1,658 (222 MiB / 27.7% CPU · 104 MiB/s · 501.4 µs CPU/req)	Not possible	1,338 (126 MiB / 27.5% CPU · 84 MiB/s · 616.2 µs CPU/req)	1,558 (130 MiB / 45.1% CPU · 97 MiB/s · 868.5 µs CPU/req)	1,003 (479 MiB / 24.2% CPU · 63 MiB/s · 722.3 µs CPU/req)
64 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	1,232 (233 MiB / 29.2% CPU · 77 MiB/s · 710.0 µs CPU/req)	🥇 1,653 (101 MiB / 22.6% CPU · 103 MiB/s · 409.2 µs CPU/req)	1,147 (129 MiB / 31.4% CPU · 72 MiB/s · 822.2 µs CPU/req)	972 (133 MiB / 45.9% CPU · 61 MiB/s · 1416.6 µs CPU/req)	550 (383 MiB / 25.7% CPU · 34 MiB/s · 1403.4 µs CPU/req)
256 KiB	HTTP/2 · plain	HTTP/1 · plain	1,124 (424 MiB / 12.3% CPU · 281 MiB/s · 329.5 µs CPU/req)	2,862 (73 MiB / 17.9% CPU · 716 MiB/s · 187.6 µs CPU/req)	2,852 (104 MiB / 24.2% CPU · 713 MiB/s · 254.1 µs CPU/req)	🥇 3,088 (145 MiB / 21.9% CPU · 772 MiB/s · 212.4 µs CPU/req)	2,097 (210 MiB / 23.4% CPU · 524 MiB/s · 335.0 µs CPU/req)
256 KiB	HTTP/2 · TLS	HTTP/2 · plain	1,355 (667 MiB / 28.7% CPU · 339 MiB/s · 634.9 µs CPU/req)	Not possible	🥇 1,522 (142 MiB / 29.7% CPU · 380 MiB/s · 585.4 µs CPU/req)	1,015 (136 MiB / 22.8% CPU · 254 MiB/s · 673.2 µs CPU/req)	1,080 (317 MiB / 29.9% CPU · 270 MiB/s · 830.7 µs CPU/req)
256 KiB	HTTP/2 · TLS	HTTP/2 · TLS	1,070 (640 MiB / 26.4% CPU · 267 MiB/s · 741.4 µs CPU/req)	Not possible	🥇 1,205 (131 MiB / 31.0% CPU · 301 MiB/s · 770.9 µs CPU/req)	1,028 (130 MiB / 24.9% CPU · 257 MiB/s · 725.9 µs CPU/req)	919 (277 MiB / 31.2% CPU · 230 MiB/s · 1018.9 µs CPU/req)
256 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	412 (499 MiB / 33.5% CPU · 103 MiB/s · 2440.4 µs CPU/req)	Not possible	🥇 528 (peak 652 · 122 MiB / 32.4% CPU · 132 MiB/s · 1841.0 µs CPU/req)	412 (139 MiB / 50.1% CPU · 103 MiB/s · 3652.9 µs CPU/req)	429 (558 MiB / 34.8% CPU · 107 MiB/s · 2432.3 µs CPU/req)
256 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	300 (peak 316 · 216 MiB / 32.6% CPU · 75 MiB/s · 3270.5 µs CPU/req)	🥇 502 (peak 532 · 98 MiB / 26.1% CPU · 125 MiB/s · 1561.3 µs CPU/req)	277 (peak 357 · 107 MiB / 25.9% CPU · 69 MiB/s · 2806.3 µs CPU/req)	485 (152 MiB / 47.4% CPU · 121 MiB/s · 2931.4 µs CPU/req)	295 (425 MiB / 30.5% CPU · 74 MiB/s · 3101.7 µs CPU/req)

Windows — POST 64 KiB request + 64 KiB response

Median of 3 repeats on windows-latest @ 482f5984. Source: Actions 37210524513 (compare-post).

Not possible: HAProxy and Envoy columns are omitted (no official Windows port).

Client	Origin	TWP	nginx	YARP
HTTP/1 · TLS	HTTP/1 · plain	🥇 5,918 (93 MiB / 44.6% CPU · 740 MiB/s · 301.5 µs CPU/req)	392 (142 MiB / 24.9% CPU · 49 MiB/s · 2538.8 µs CPU/req)	4,163 (138 MiB / 53.9% CPU · 520 MiB/s · 518.2 µs CPU/req)
HTTP/2 · TLS	HTTP/1 · plain	🥇 4,206 (216 MiB / 46.9% CPU · 526 MiB/s · 446.5 µs CPU/req)	359 (147 MiB / 24.4% CPU · 45 MiB/s · 2722.1 µs CPU/req)	4,053 (140 MiB / 47.7% CPU · 507 MiB/s · 471.0 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · plain	🥇 2,848 (202 MiB / 46.7% CPU · 356 MiB/s · 656.4 µs CPU/req)	Not possible (no QUIC)	1,628 (218 MiB / 48.1% CPU · 203 MiB/s · 1180.9 µs CPU/req)
HTTP/2 · plain	HTTP/1 · plain	🥇 8,704 (211 MiB / 45.0% CPU · 1,088 MiB/s · 206.9 µs CPU/req)	3,912 (132 MiB / 24.8% CPU · 489 MiB/s · 253.6 µs CPU/req)	8,145 (120 MiB / 48.4% CPU · 1,018 MiB/s · 237.8 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · plain	🥇 4,553 (169 MiB / 40.4% CPU · 569 MiB/s · 355.3 µs CPU/req)	Not possible	3,588 (142 MiB / 46.4% CPU · 448 MiB/s · 517.7 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · TLS	🥇 3,570 (171 MiB / 39.1% CPU · 446 MiB/s · 437.9 µs CPU/req)	Not possible	2,886 (157 MiB / 47.5% CPU · 361 MiB/s · 658.9 µs CPU/req)
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 1,889 (190 MiB / 44.2% CPU · 236 MiB/s · 936.7 µs CPU/req)	Not possible	1,135 (225 MiB / 47.8% CPU · 142 MiB/s · 1684.2 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 1,908 (192 MiB / 46.7% CPU · 239 MiB/s · 978.5 µs CPU/req)	Not possible (no QUIC)	1,292 (219 MiB / 50.7% CPU · 161 MiB/s · 1570.3 µs CPU/req)

TWP leads H2→H1 POST (about 1.1–1.3× YARP) and H3 POST (about 1.0–1.1× YARP). H2 TLS→H2 TLS POST sustain is about 1.2–1.4× YARP.

Linux — POST 64 KiB request + 64 KiB response

Median of 3 repeats @ 482f5984. Source: Actions 37210524513 (compare-post).

Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
HTTP/1 · TLS	HTTP/1 · plain	10,424 (133 MiB / 39.9% CPU · 1,303 MiB/s · 153.0 µs CPU/req)	7,771 (100 MiB / 48.2% CPU · 971 MiB/s · 248.1 µs CPU/req)	🥇 11,193 (88 MiB / 39.3% CPU · 1,399 MiB/s · 140.3 µs CPU/req)	10,756 (133 MiB / 39.1% CPU · 1,344 MiB/s · 145.5 µs CPU/req)	6,818 (170 MiB / 53.0% CPU · 852 MiB/s · 310.9 µs CPU/req)
HTTP/2 · TLS	HTTP/1 · plain	3,766 (252 MiB / 47.6% CPU · 471 MiB/s · 506.1 µs CPU/req)	3,294 (106 MiB / 46.1% CPU · 412 MiB/s · 560.2 µs CPU/req)	🥇 3,941 (91 MiB / 44.6% CPU · 493 MiB/s · 452.4 µs CPU/req)	3,640 (peak 4,454 · 126 MiB / 42.0% CPU · 455 MiB/s · 461.1 µs CPU/req)	3,380 (169 MiB / 50.2% CPU · 423 MiB/s · 593.7 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · plain	🥇 3,008 (244 MiB / 45.2% CPU · 376 MiB/s · 601.7 µs CPU/req)	968 (111 MiB / 42.4% CPU · 121 MiB/s · 1753.1 µs CPU/req)	2,580 (93 MiB / 42.8% CPU · 323 MiB/s · 664.0 µs CPU/req)	0 (130 MiB / 0.3% CPU)	2,178 (248 MiB / 50.3% CPU · 272 MiB/s · 924.2 µs CPU/req)
HTTP/2 · plain	HTTP/1 · plain	5,843 (275 MiB / 44.1% CPU · 730 MiB/s · 302.2 µs CPU/req)	4,768 (85 MiB / 45.0% CPU · 596 MiB/s · 377.3 µs CPU/req)	🥇 6,440 (74 MiB / 40.1% CPU · 805 MiB/s · 249.3 µs CPU/req)	4,368 (peak 5,801 · 116 MiB / 44.4% CPU · 546 MiB/s · 406.7 µs CPU/req)	5,532 (160 MiB / 47.9% CPU · 692 MiB/s · 346.5 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · plain	🥇 3,960 (204 MiB / 39.8% CPU · 495 MiB/s · 401.9 µs CPU/req)	Not possible	3,326 (93 MiB / 42.0% CPU · 416 MiB/s · 504.6 µs CPU/req)	3,914 (136 MiB / 40.6% CPU · 489 MiB/s · 414.8 µs CPU/req)	2,662 (189 MiB / 49.1% CPU · 333 MiB/s · 738.1 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · TLS	🥇 5,239 (214 MiB / 37.6% CPU · 655 MiB/s · 287.1 µs CPU/req)	Not possible	4,683 (91 MiB / 37.9% CPU · 585 MiB/s · 323.5 µs CPU/req)	5,172 (141 MiB / 39.4% CPU · 647 MiB/s · 304.5 µs CPU/req)	3,632 (197 MiB / 47.1% CPU · 454 MiB/s · 518.6 µs CPU/req)
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 1,989 (242 MiB / 44.9% CPU · 249 MiB/s · 902.9 µs CPU/req)	Not possible	1,768 (96 MiB / 44.0% CPU · 221 MiB/s · 995.8 µs CPU/req)	0 (130 MiB / 0.2% CPU)	1,532 (253 MiB / 48.6% CPU · 191 MiB/s · 1270.1 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 2,230 (249 MiB / 46.5% CPU · 279 MiB/s · 834.7 µs CPU/req)	872 (113 MiB / 52.1% CPU · 109 MiB/s · 2392.8 µs CPU/req)	1,994 (96 MiB / 44.3% CPU · 249 MiB/s · 888.5 µs CPU/req)	0 (132 MiB / 0.3% CPU)	1,797 (276 MiB / 49.4% CPU · 225 MiB/s · 1099.6 µs CPU/req)

Linux nginx H1/H2/H3 POST completed (nginx.org mainline). TWP÷YARP H1 ≈ 1.5×; H2→H1 ≈ 1.2×; H3 ≈ 1.1×. TWP÷nginx H3 ≈ 6×.

macOS — POST 64 KiB request + 64 KiB response

Median of 3 repeats on macos-15 (Apple Silicon M1, 3-core / 7 GB) @ 482f5984. Source: Actions 37210524513 (compare-post).

Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
HTTP/1 · TLS	HTTP/1 · plain	2,466 (250 MiB / 21.9% CPU · 308 MiB/s · 267.0 µs CPU/req)	🥇 3,778 (88 MiB / 20.1% CPU · 472 MiB/s · 159.7 µs CPU/req)	3,778 (120 MiB / 26.9% CPU · 472 MiB/s · 213.4 µs CPU/req)	3,534 (131 MiB / 25.9% CPU · 442 MiB/s · 219.9 µs CPU/req)	1,959 (212 MiB / 30.1% CPU · 245 MiB/s · 461.2 µs CPU/req)
HTTP/2 · TLS	HTTP/1 · plain	2,035 (550 MiB / 24.9% CPU · 254 MiB/s · 367.7 µs CPU/req)	2,606 (99 MiB / 18.9% CPU · 326 MiB/s · 218.0 µs CPU/req)	🥇 2,857 (124 MiB / 29.5% CPU · 357 MiB/s · 309.7 µs CPU/req)	1,932 (peak 2,713 · 123 MiB / 19.4% CPU · 241 MiB/s · 300.7 µs CPU/req)	1,679 (186 MiB / 24.4% CPU · 210 MiB/s · 435.2 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · plain	1,402 (408 MiB / 28.3% CPU · 175 MiB/s · 605.3 µs CPU/req)	🥇 1,842 (97 MiB / 24.3% CPU · 230 MiB/s · 395.6 µs CPU/req)	961 (102 MiB / 24.2% CPU · 120 MiB/s · 756.9 µs CPU/req)	1,122 (130 MiB / 42.4% CPU · 140 MiB/s · 1133.2 µs CPU/req)	1,330 (469 MiB / 30.2% CPU · 166 MiB/s · 682.5 µs CPU/req)
HTTP/2 · plain	HTTP/1 · plain	2,022 (528 MiB / 19.7% CPU · 253 MiB/s · 291.8 µs CPU/req)	🥇 6,081 (86 MiB / 17.7% CPU · 760 MiB/s · 87.1 µs CPU/req)	4,151 (115 MiB / 22.8% CPU · 519 MiB/s · 164.6 µs CPU/req)	4,234 (136 MiB / 19.5% CPU · 529 MiB/s · 138.2 µs CPU/req)	2,082 (215 MiB / 19.1% CPU · 260 MiB/s · 274.7 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · plain	2,391 (413 MiB / 17.3% CPU · 299 MiB/s · 216.5 µs CPU/req)	Not possible	🥇 2,970 (128 MiB / 28.5% CPU · 371 MiB/s · 287.9 µs CPU/req)	2,717 (128 MiB / 23.4% CPU · 340 MiB/s · 258.1 µs CPU/req)	997 (230 MiB / 16.8% CPU · 125 MiB/s · 504.6 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · TLS	1,866 (449 MiB / 24.3% CPU · 233 MiB/s · 390.2 µs CPU/req)	Not possible	🥇 2,847 (130 MiB / 32.1% CPU · 356 MiB/s · 338.2 µs CPU/req)	2,722 (128 MiB / 27.2% CPU · 340 MiB/s · 299.5 µs CPU/req)	1,261 (221 MiB / 25.1% CPU · 158 MiB/s · 595.9 µs CPU/req)
HTTP/3 · QUIC	HTTP/2 · TLS	663 (241 MiB / 26.9% CPU · 83 MiB/s · 1218.0 µs CPU/req)	Not possible	🥇 676 (123 MiB / 32.5% CPU · 84 MiB/s · 1441.2 µs CPU/req)	603 (137 MiB / 45.2% CPU · 75 MiB/s · 2249.0 µs CPU/req)	610 (456 MiB / 26.8% CPU · 76 MiB/s · 1319.1 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · TLS	1,048 (309 MiB / 30.1% CPU · 131 MiB/s · 860.6 µs CPU/req)	🥇 1,060 (102 MiB / 20.8% CPU · 133 MiB/s · 589.1 µs CPU/req)	867 (124 MiB / 30.4% CPU · 108 MiB/s · 1052.0 µs CPU/req)	474 (140 MiB / 28.5% CPU · 59 MiB/s · 1803.8 µs CPU/req)	983 (420 MiB / 33.0% CPU · 123 MiB/s · 1005.8 µs CPU/req)

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

Userspace 5 ms one-way delay + 1% TCP connection stall (H1/H2); UDP is loss% only (no per-datagram delay; MsQuic-safe) + 1% datagram drop (H3); 64 KiB GET. Median of 3 repeats on windows-latest @ 482f5984 — 37210527573 (compare-lossy).

Not possible: HAProxy and Envoy columns are omitted (no official Windows port).

Client	Origin	TWP	nginx	YARP
HTTP/1 · TLS	HTTP/1 · plain	🥇 664 (85 MiB / 4.2% CPU · 41 MiB/s · 252.5 µs CPU/req)	639 (143 MiB / 19.3% CPU · 40 MiB/s · 1208.4 µs CPU/req)	661 (122 MiB / 5.6% CPU · 41 MiB/s · 337.6 µs CPU/req)
HTTP/2 · TLS	HTTP/1 · plain	🥇 248 (150 MiB / 2.6% CPU · 15 MiB/s · 418.4 µs CPU/req)	245 (143 MiB / 6.3% CPU · 15 MiB/s · 1033.5 µs CPU/req)	234 (119 MiB / 2.5% CPU · 15 MiB/s · 430.0 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · plain	836 (123 MiB / 18.2% CPU · 52 MiB/s · 869.3 µs CPU/req)	Not possible (no QUIC)	🥇 841 (174 MiB / 31.1% CPU · 53 MiB/s · 1478.5 µs CPU/req)
HTTP/2 · plain	HTTP/1 · plain	🥇 253 (153 MiB / 2.7% CPU · 16 MiB/s · 425.0 µs CPU/req)	250 (128 MiB / 2.4% CPU · 16 MiB/s · 383.0 µs CPU/req)	231 (106 MiB / 2.4% CPU · 14 MiB/s · 414.9 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · plain	220 (122 MiB / 2.1% CPU · 14 MiB/s · 382.5 µs CPU/req)	Not possible	🥇 234 (129 MiB / 3.4% CPU · 15 MiB/s · 580.7 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · TLS	🥇 231 (131 MiB / 2.9% CPU · 14 MiB/s · 499.6 µs CPU/req)	Not possible	229 (114 MiB / 4.7% CPU · 14 MiB/s · 827.6 µs CPU/req)
HTTP/3 · QUIC	HTTP/2 · TLS	893 (116 MiB / 20.6% CPU · 56 MiB/s · 922.1 µs CPU/req)	Not possible	🥇 912 (185 MiB / 31.2% CPU · 57 MiB/s · 1368.3 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 851 (131 MiB / 20.1% CPU · 53 MiB/s · 945.8 µs CPU/req)	Not possible (no QUIC)	844 (177 MiB / 31.1% CPU · 53 MiB/s · 1474.8 µs CPU/req)

H1, H2, and H3 loss sustain are in the table. Prefer the ratio columns.

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

Median of 3 repeats @ 482f5984. Source: 37210527573 (compare-lossy; lossy H3 uses quic-http3, UDP drop-only).

Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
HTTP/1 · TLS	HTTP/1 · plain	1,200 (121 MiB / 13.1% CPU · 75 MiB/s · 437.9 µs CPU/req)	🥇 1,209 (99 MiB / 12.7% CPU · 76 MiB/s · 420.8 µs CPU/req)	1,205 (84 MiB / 6.8% CPU · 75 MiB/s · 225.8 µs CPU/req)	1,195 (129 MiB / 9.1% CPU · 75 MiB/s · 304.2 µs CPU/req)	1,197 (155 MiB / 16.4% CPU · 75 MiB/s · 548.8 µs CPU/req)
HTTP/2 · TLS	HTTP/1 · plain	🥇 548 (189 MiB / 11.7% CPU · 34 MiB/s · 855.4 µs CPU/req)	507 (99 MiB / 5.1% CPU · 32 MiB/s · 405.4 µs CPU/req)	511 (84 MiB / 4.1% CPU · 32 MiB/s · 320.9 µs CPU/req)	496 (130 MiB / 4.9% CPU · 31 MiB/s · 393.7 µs CPU/req)	474 (143 MiB / 12.3% CPU · 30 MiB/s · 1033.9 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · plain	🥇 1,195 (163 MiB / 24.2% CPU · 75 MiB/s · 810.0 µs CPU/req)	1,096 (113 MiB / 23.5% CPU · 68 MiB/s · 857.1 µs CPU/req)	978 (101 MiB / 18.8% CPU · 61 MiB/s · 769.4 µs CPU/req)	1,022 (138 MiB / 25.8% CPU · 64 MiB/s · 1007.9 µs CPU/req)	1,144 (231 MiB / 31.2% CPU · 72 MiB/s · 1090.8 µs CPU/req)
HTTP/2 · plain	HTTP/1 · plain	🥇 619 (186 MiB / 10.0% CPU · 39 MiB/s · 646.0 µs CPU/req)	552 (78 MiB / 3.4% CPU · 34 MiB/s · 248.0 µs CPU/req)	533 (71 MiB / 2.4% CPU · 33 MiB/s · 177.1 µs CPU/req)	521 (122 MiB / 3.1% CPU · 33 MiB/s · 241.7 µs CPU/req)	474 (133 MiB / 8.5% CPU · 30 MiB/s · 720.1 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · plain	397 (148 MiB / 7.7% CPU · 25 MiB/s · 771.4 µs CPU/req)	Not possible	🥇 532 (86 MiB / 3.9% CPU · 33 MiB/s · 291.9 µs CPU/req)	506 (130 MiB / 4.8% CPU · 32 MiB/s · 375.7 µs CPU/req)	415 (148 MiB / 11.7% CPU · 26 MiB/s · 1125.1 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · TLS	286 (161 MiB / 4.0% CPU · 18 MiB/s · 561.6 µs CPU/req)	Not possible	🥇 552 (88 MiB / 2.6% CPU · 34 MiB/s · 187.7 µs CPU/req)	512 (132 MiB / 2.9% CPU · 32 MiB/s · 224.3 µs CPU/req)	389 (148 MiB / 6.5% CPU · 24 MiB/s · 669.6 µs CPU/req)
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 1,146 (160 MiB / 25.7% CPU · 72 MiB/s · 896.6 µs CPU/req)	Not possible	921 (105 MiB / 20.5% CPU · 58 MiB/s · 891.9 µs CPU/req)	957 (141 MiB / 26.4% CPU · 60 MiB/s · 1103.9 µs CPU/req)	1,055 (229 MiB / 32.7% CPU · 66 MiB/s · 1238.8 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 1,102 (179 MiB / 26.1% CPU · 69 MiB/s · 945.9 µs CPU/req)	1,015 (117 MiB / 24.6% CPU · 63 MiB/s · 967.1 µs CPU/req)	947 (103 MiB / 20.3% CPU · 59 MiB/s · 857.0 µs CPU/req)	996 (137 MiB / 25.9% CPU · 62 MiB/s · 1042.7 µs CPU/req)	1,048 (219 MiB / 32.9% CPU · 66 MiB/s · 1256.0 µs CPU/req)

TWP H2 HOL ≫ YARP (~7.8×). H3 TWP÷YARP ≈ 1.07×.

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

Median of 3 repeats on macos-15 (Apple Silicon M1, 3-core / 7 GB) @ 482f5984. Source: 37210527573 (compare-lossy; lossy H3 uses quic-http3, UDP drop-only).

Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
HTTP/1 · TLS	HTTP/1 · plain	366 (233 MiB / 4.3% CPU · 23 MiB/s · 351.7 µs CPU/req)	🥇 475 (85 MiB / 1.6% CPU · 30 MiB/s · 99.7 µs CPU/req)	399 (115 MiB / 1.6% CPU · 25 MiB/s · 124.0 µs CPU/req)	443 (129 MiB / 2.9% CPU · 28 MiB/s · 195.1 µs CPU/req)	393 (226 MiB / 7.3% CPU · 25 MiB/s · 556.6 µs CPU/req)
HTTP/2 · TLS	HTTP/1 · plain	🥇 161 (414 MiB / 4.5% CPU · 10 MiB/s · 845.7 µs CPU/req)	143 (85 MiB / 0.8% CPU · 9.0 MiB/s · 156.9 µs CPU/req)	144 (116 MiB / 1.4% CPU · 9.0 MiB/s · 286.4 µs CPU/req)	120 (130 MiB / 1.2% CPU · 7.5 MiB/s · 290.2 µs CPU/req)	112 (230 MiB / 2.2% CPU · 7.0 MiB/s · 601.6 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · plain	1,052 (267 MiB / 19.7% CPU · 66 MiB/s · 561.2 µs CPU/req)	203 (peak 358 · 81 MiB / 3.7% CPU · 13 MiB/s · 544.5 µs CPU/req)	728 (123 MiB / 15.0% CPU · 46 MiB/s · 618.0 µs CPU/req)	🥇 1,216 (132 MiB / 28.4% CPU · 76 MiB/s · 700.2 µs CPU/req)	960 (408 MiB / 22.3% CPU · 60 MiB/s · 698.2 µs CPU/req)
HTTP/2 · plain	HTTP/1 · plain	144 (357 MiB / 2.7% CPU · 9.0 MiB/s · 571.6 µs CPU/req)	136 (73 MiB / 0.4% CPU · 8.5 MiB/s · 92.9 µs CPU/req)	143 (104 MiB / 0.7% CPU · 8.9 MiB/s · 148.8 µs CPU/req)	128 (129 MiB / 0.6% CPU · 8.0 MiB/s · 143.1 µs CPU/req)	🥇 144 (144 MiB / 2.0% CPU · 9.0 MiB/s · 423.2 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · plain	136 (297 MiB / 1.9% CPU · 8.5 MiB/s · 431.1 µs CPU/req)	Not possible	136 (121 MiB / 0.8% CPU · 8.5 MiB/s · 181.0 µs CPU/req)	121 (123 MiB / 1.4% CPU · 7.6 MiB/s · 353.4 µs CPU/req)	🥇 159 (277 MiB / 7.8% CPU · 9.9 MiB/s · 1470.7 µs CPU/req)
HTTP/2 · TLS	HTTP/2 · TLS	176 (313 MiB / 7.3% CPU · 11 MiB/s · 1238.9 µs CPU/req)	Not possible	🥇 179 (119 MiB / 3.1% CPU · 11 MiB/s · 521.8 µs CPU/req)	151 (124 MiB / 2.8% CPU · 9.5 MiB/s · 556.8 µs CPU/req)	168 (247 MiB / 6.2% CPU · 10 MiB/s · 1107.3 µs CPU/req)
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 968 (199 MiB / 24.5% CPU · 60 MiB/s · 758.5 µs CPU/req)	Not possible	442 (132 MiB / 16.2% CPU · 28 MiB/s · 1099.9 µs CPU/req)	768 (131 MiB / 27.4% CPU · 48 MiB/s · 1071.2 µs CPU/req)	477 (433 MiB / 19.2% CPU · 30 MiB/s · 1206.8 µs CPU/req)
HTTP/3 · QUIC	HTTP/1 · TLS	880 (210 MiB / 21.1% CPU · 55 MiB/s · 718.5 µs CPU/req)	🥇 1,004 (99 MiB / 13.5% CPU · 63 MiB/s · 402.3 µs CPU/req)	436 (126 MiB / 14.2% CPU · 27 MiB/s · 976.4 µs CPU/req)	952 (132 MiB / 27.3% CPU · 60 MiB/s · 859.7 µs CPU/req)	545 (356 MiB / 22.6% CPU · 34 MiB/s · 1245.9 µs CPU/req)

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 @ 482f5984 (37210530448). 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 row = H1 Upgrade echo round-trips/sec (not RFC 8441; see WebSocket RFC 8441 table below).

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

Windows

Not possible: HAProxy and Envoy columns are omitted (no official Windows port).

Scenario	Client	Origin	TWP	nginx	YARP
Slow consumer (256 KiB GET, throttled client read)	HTTP/1 · TLS	HTTP/1 · plain	250 (95 MiB / 3.0% CPU · 63 MiB/s · 479.6 µs CPU/req)	242 (144 MiB / 16.1% CPU · 60 MiB/s · 2663.9 µs CPU/req)	🥇 256 (110 MiB / 3.0% CPU · 64 MiB/s · 477.1 µs CPU/req)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/1 · plain	248 (143 MiB / 6.0% CPU · 62 MiB/s · 970.5 µs CPU/req)	158 (142 MiB / 24.8% CPU · 40 MiB/s · 6268.8 µs CPU/req)	🥇 248 (121 MiB / 6.7% CPU · 62 MiB/s · 1086.4 µs CPU/req)
Slow consumer (256 KiB GET, throttled client read)	HTTP/3 · QUIC	HTTP/1 · plain	64 (108 MiB / 46.1% CPU · 16 MiB/s · 28974.8 µs CPU/req)	Not possible (no QUIC)	🥇 86 (153 MiB / 47.5% CPU · 22 MiB/s · 21972.2 µs CPU/req)
Early response (origin writes after first request chunk)	HTTP/1 · TLS	HTTP/1 · plain	🥇 4,529 (101 MiB / 35.5% CPU · 566 MiB/s · 313.2 µs CPU/req)	341 (143 MiB / 24.8% CPU · 43 MiB/s · 2903.3 µs CPU/req)	4,090 (139 MiB / 56.7% CPU · 511 MiB/s · 554.5 µs CPU/req)
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/1 · plain	5,802 (213 MiB / 49.5% CPU · 725 MiB/s · 341.1 µs CPU/req)	653 (146 MiB / 25.0% CPU · 82 MiB/s · 1530.0 µs CPU/req)	🥇 7,528 (138 MiB / 46.5% CPU · 941 MiB/s · 247.2 µs CPU/req)
Early response (origin writes after first request chunk)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 1,677 (163 MiB / 44.8% CPU · 210 MiB/s · 1068.6 µs CPU/req)	Not possible (no QUIC)	1,013 (194 MiB / 49.4% CPU · 127 MiB/s · 1951.5 µs CPU/req)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · plain	HTTP/1 · plain	248 (132 MiB / 3.4% CPU · 62 MiB/s · 555.7 µs CPU/req)	248 (128 MiB / 7.9% CPU · 62 MiB/s · 1271.4 µs CPU/req)	🥇 248 (110 MiB / 3.8% CPU · 62 MiB/s · 616.9 µs CPU/req)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/2 · TLS	243 (160 MiB / 8.9% CPU · 61 MiB/s · 1468.3 µs CPU/req)	Not possible	🥇 248 (146 MiB / 8.9% CPU · 62 MiB/s · 1431.9 µs CPU/req)
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/2 · TLS	🥇 3,292 (196 MiB / 41.2% CPU · 412 MiB/s · 500.1 µs CPU/req)	Not possible	2,707 (162 MiB / 54.6% CPU · 338 MiB/s · 806.2 µs CPU/req)
Duplex (both directions live)	HTTP/2 · TLS	HTTP/2 · TLS	🥇 6,092 (191 MiB / 40.2% CPU · 762 MiB/s · 263.8 µs CPU/req)	Not possible	4,571 (172 MiB / 53.6% CPU · 571 MiB/s · 468.7 µs CPU/req)
Duplex (WebSocket / H1 Upgrade)	HTTP/1 · TLS	HTTP/1 · plain	🥇 25,553 (97 MiB / 42.8% CPU · 67.0 µs CPU/req)	12,474 (143 MiB / 24.7% CPU · 79.1 µs CPU/req)	23,549 (89 MiB / 42.0% CPU · 71.3 µs CPU/req)

Linux

Scenario	Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
Slow consumer (256 KiB GET, throttled client read)	HTTP/1 · TLS	HTTP/1 · plain	466 (118 MiB / 6.5% CPU · 117 MiB/s · 559.5 µs CPU/req)	424 (101 MiB / 5.9% CPU · 106 MiB/s · 555.1 µs CPU/req)	472 (85 MiB / 3.6% CPU · 118 MiB/s · 303.3 µs CPU/req)	🥇 472 (139 MiB / 5.1% CPU · 118 MiB/s · 431.1 µs CPU/req)	405 (147 MiB / 9.9% CPU · 101 MiB/s · 974.1 µs CPU/req)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/1 · plain	477 (167 MiB / 15.2% CPU · 119 MiB/s · 1269.6 µs CPU/req)	476 (101 MiB / 10.1% CPU · 119 MiB/s · 848.8 µs CPU/req)	🥇 479 (83 MiB / 7.0% CPU · 120 MiB/s · 581.8 µs CPU/req)	478 (143 MiB / 6.9% CPU · 120 MiB/s · 575.4 µs CPU/req)	474 (158 MiB / 16.6% CPU · 118 MiB/s · 1401.7 µs CPU/req)
Slow consumer (256 KiB GET, throttled client read)	HTTP/3 · QUIC	HTTP/1 · plain	462 (147 MiB / 48.6% CPU · 115 MiB/s · 4208.8 µs CPU/req)	462 (112 MiB / 26.6% CPU · 116 MiB/s · 2300.3 µs CPU/req)	🥇 472 (92 MiB / 13.7% CPU · 118 MiB/s · 1158.8 µs CPU/req)	461 (150 MiB / 24.9% CPU · 115 MiB/s · 2160.1 µs CPU/req)	454 (199 MiB / 50.6% CPU · 114 MiB/s · 4456.3 µs CPU/req)
Early response (origin writes after first request chunk)	HTTP/1 · TLS	HTTP/1 · plain	9,667 (132 MiB / 43.1% CPU · 1,208 MiB/s · 178.2 µs CPU/req)	7,106 (peak 7,239 · 101 MiB / 49.9% CPU · 888 MiB/s · 281.1 µs CPU/req)	🥇 10,897 (90 MiB / 42.9% CPU · 1,362 MiB/s · 157.6 µs CPU/req)	0 (peak 64 · 130 MiB / 1.8% CPU)	6,468 (175 MiB / 54.2% CPU · 809 MiB/s · 335.0 µs CPU/req)
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/1 · plain	3,036 (272 MiB / 52.8% CPU · 379 MiB/s · 695.4 µs CPU/req)	3,219 (106 MiB / 45.4% CPU · 402 MiB/s · 564.0 µs CPU/req)	🥇 3,862 (91 MiB / 43.6% CPU · 483 MiB/s · 451.3 µs CPU/req)	0 (peak 3,872 · 140 MiB / 42.3% CPU)	3,306 (169 MiB / 50.1% CPU · 413 MiB/s · 606.7 µs CPU/req)
Early response (origin writes after first request chunk)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 3,355 (254 MiB / 45.1% CPU · 419 MiB/s · 537.7 µs CPU/req)	1,554 (116 MiB / 29.2% CPU · 194 MiB/s · 751.1 µs CPU/req)	3,056 (96 MiB / 37.6% CPU · 382 MiB/s · 491.9 µs CPU/req)	0 (127 MiB / 0.1% CPU)	2,201 (287 MiB / 48.0% CPU · 275 MiB/s · 872.2 µs CPU/req)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · plain	HTTP/1 · plain	🥇 480 (157 MiB / 17.2% CPU · 120 MiB/s · 1434.5 µs CPU/req)	478 (77 MiB / 14.0% CPU · 119 MiB/s · 1173.6 µs CPU/req)	477 (70 MiB / 7.3% CPU · 119 MiB/s · 613.8 µs CPU/req)	472 (134 MiB / 7.3% CPU · 118 MiB/s · 615.9 µs CPU/req)	478 (141 MiB / 16.6% CPU · 119 MiB/s · 1393.1 µs CPU/req)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/2 · TLS	463 (189 MiB / 23.9% CPU · 116 MiB/s · 2065.6 µs CPU/req)	Not possible	469 (95 MiB / 19.9% CPU · 117 MiB/s · 1696.7 µs CPU/req)	🥇 469 (140 MiB / 19.0% CPU · 117 MiB/s · 1621.1 µs CPU/req)	467 (170 MiB / 32.2% CPU · 117 MiB/s · 2764.7 µs CPU/req)
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/2 · TLS	🥇 2,854 (228 MiB / 41.6% CPU · 357 MiB/s · 582.5 µs CPU/req)	Not possible	2,039 (93 MiB / 34.4% CPU · 255 MiB/s · 674.8 µs CPU/req)	0 (peak 2,457 · 142 MiB / 35.5% CPU)	1,955 (201 MiB / 53.9% CPU · 244 MiB/s · 1102.0 µs CPU/req)
Duplex (both directions live)	HTTP/2 · TLS	HTTP/2 · TLS	🥇 4,025 (227 MiB / 38.7% CPU · 503 MiB/s · 384.5 µs CPU/req)	Not possible	2,873 (93 MiB / 33.9% CPU · 359 MiB/s · 471.3 µs CPU/req)	0 (peak 3,222 · 144 MiB / 36.0% CPU)	2,870 (214 MiB / 54.2% CPU · 359 MiB/s · 755.0 µs CPU/req)
Duplex (WebSocket / H1 Upgrade)	HTTP/1 · TLS	HTTP/1 · plain	29,881 (126 MiB / 43.6% CPU · 58.4 µs CPU/req)	🥇 33,514 (99 MiB / 36.1% CPU · 43.1 µs CPU/req)	32,603 (83 MiB / 39.0% CPU · 47.8 µs CPU/req)	31,792 (127 MiB / 39.8% CPU · 50.1 µs CPU/req)	27,275 (126 MiB / 44.1% CPU · 64.7 µs CPU/req)

macOS

Median of 3 repeats on macos-15 (Apple Silicon M1, 3-core / 7 GB) @ 482f5984. Source: Actions 37210530448 (compare-arch).

Scenario	Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
Slow consumer (256 KiB GET, throttled client read)	HTTP/1 · TLS	HTTP/1 · plain	96 (246 MiB / 2.4% CPU · 24 MiB/s · 748.4 µs CPU/req)	103 (84 MiB / 1.0% CPU · 26 MiB/s · 291.8 µs CPU/req)	🥇 108 (111 MiB / 1.1% CPU · 27 MiB/s · 306.9 µs CPU/req)	90 (150 MiB / 1.3% CPU · 23 MiB/s · 438.5 µs CPU/req)	88 (205 MiB / 2.9% CPU · 22 MiB/s · 984.1 µs CPU/req)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/1 · plain	96 (370 MiB / 4.1% CPU · 24 MiB/s · 1299.2 µs CPU/req)	96 (84 MiB / 1.8% CPU · 24 MiB/s · 547.4 µs CPU/req)	🥇 112 (114 MiB / 2.2% CPU · 28 MiB/s · 584.5 µs CPU/req)	88 (153 MiB / 1.9% CPU · 22 MiB/s · 667.0 µs CPU/req)	88 (221 MiB / 4.3% CPU · 22 MiB/s · 1462.4 µs CPU/req)
Slow consumer (256 KiB GET, throttled client read)	HTTP/3 · QUIC	HTTP/1 · plain	71 (132 MiB / 19.2% CPU · 18 MiB/s · 8111.0 µs CPU/req)	104 (94 MiB / 4.6% CPU · 26 MiB/s · 1332.4 µs CPU/req)	🥇 120 (110 MiB / 12.6% CPU · 30 MiB/s · 3163.0 µs CPU/req)	88 (149 MiB / 10.8% CPU · 22 MiB/s · 3679.2 µs CPU/req)	113 (333 MiB / 21.1% CPU · 28 MiB/s · 5597.2 µs CPU/req)
Early response (origin writes after first request chunk)	HTTP/1 · TLS	HTTP/1 · plain	4,028 (238 MiB / 25.5% CPU · 504 MiB/s · 190.1 µs CPU/req)	2,920 (peak 3,205 · 85 MiB / 17.1% CPU · 365 MiB/s · 176.1 µs CPU/req)	🥇 4,105 (120 MiB / 29.0% CPU · 513 MiB/s · 211.7 µs CPU/req)	0 (peak 3,376 · 134 MiB / 26.5% CPU)	1,911 (203 MiB / 28.4% CPU · 239 MiB/s · 446.2 µs CPU/req)
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/1 · plain	1,631 (514 MiB / 30.5% CPU · 204 MiB/s · 561.6 µs CPU/req)	2,380 (peak 3,110 · 98 MiB / 16.4% CPU · 297 MiB/s · 207.1 µs CPU/req)	🥇 3,369 (122 MiB / 27.7% CPU · 421 MiB/s · 247.0 µs CPU/req)	0 (peak 3,882 · 144 MiB / 27.4% CPU)	1,801 (192 MiB / 20.8% CPU · 225 MiB/s · 346.2 µs CPU/req)
Early response (origin writes after first request chunk)	HTTP/3 · QUIC	HTTP/1 · plain	744 (255 MiB / 26.4% CPU · 93 MiB/s · 1065.8 µs CPU/req)	🥇 1,512 (100 MiB / 22.1% CPU · 189 MiB/s · 438.4 µs CPU/req)	0 (65 MiB)	542 (125 MiB / 30.7% CPU · 68 MiB/s · 1699.6 µs CPU/req)	408 (436 MiB / 23.6% CPU · 51 MiB/s · 1734.3 µs CPU/req)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · plain	HTTP/1 · plain	96 (208 MiB / 2.1% CPU · 24 MiB/s · 651.4 µs CPU/req)	🥇 120 (72 MiB / 1.9% CPU · 30 MiB/s · 478.3 µs CPU/req)	104 (99 MiB / 1.2% CPU · 26 MiB/s · 349.4 µs CPU/req)	81 (145 MiB / 1.0% CPU · 20 MiB/s · 383.4 µs CPU/req)	104 (145 MiB / 3.2% CPU · 26 MiB/s · 922.0 µs CPU/req)
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/2 · TLS	🥇 104 (481 MiB / 2.8% CPU · 26 MiB/s · 805.8 µs CPU/req)	Not possible	96 (130 MiB / 2.0% CPU · 24 MiB/s · 633.9 µs CPU/req)	80 (130 MiB / 1.5% CPU · 20 MiB/s · 554.0 µs CPU/req)	96 (347 MiB / 3.2% CPU · 24 MiB/s · 987.5 µs CPU/req)
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/2 · TLS	1,546 (446 MiB / 22.0% CPU · 193 MiB/s · 426.1 µs CPU/req)	Not possible	1,609 (119 MiB / 20.9% CPU · 201 MiB/s · 389.2 µs CPU/req)	0 (peak 1,610 · 134 MiB / 24.2% CPU)	🥇 1,645 (349 MiB / 30.2% CPU · 206 MiB/s · 550.9 µs CPU/req)
Duplex (both directions live)	HTTP/2 · TLS	HTTP/2 · TLS	1,812 (460 MiB / 17.6% CPU · 227 MiB/s · 291.5 µs CPU/req)	Not possible	🥇 1,945 (peak 2,028 · 119 MiB / 17.9% CPU · 243 MiB/s · 275.3 µs CPU/req)	0 (peak 1,669 · 131 MiB / 21.3% CPU)	1,677 (368 MiB / 31.5% CPU · 210 MiB/s · 563.4 µs CPU/req)
Duplex (WebSocket / H1 Upgrade)	HTTP/1 · TLS	HTTP/1 · plain	5,470 (156 MiB / 10.0% CPU · 54.7 µs CPU/req)	🥇 17,987 (86 MiB / 15.6% CPU · 26.1 µs CPU/req)	10,901 (103 MiB / 13.7% CPU · 37.8 µs CPU/req)	12,646 (121 MiB / 21.1% CPU · 50.1 µs CPU/req)	15,211 (152 MiB / 21.5% CPU · 42.5 µs CPU/req)

Slow consumer is sleep-bound; H1/H2/H3 sit in the same band. Prefer the ratio columns for early-response, duplex, and WebSocket rows.

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 @ 482f5984. Source: Actions 37210533058. Absolute RPS on GHA swings hard; prefer TWP÷YARP.

Not possible: HAProxy and Envoy columns are omitted (no official Windows port).

Workload	TWP	nginx	YARP
Keep-alive · tiny GET	🥇 55,831 (85 MiB / 47.4% CPU · 34.0 µs CPU/req)	24,990 (142 MiB / 24.4% CPU · 39.1 µs CPU/req)	52,516 (105 MiB / 49.0% CPU · 37.3 µs CPU/req)
New-connection · tiny GET	🥇 928 (78 MiB)	319 (141 MiB / 1.9% CPU · 234.8 µs CPU/req)	920 (105 MiB)
Keep-alive · 256 KiB GET	🥇 5,229 (128 MiB / 37.9% CPU · 1,307 MiB/s · 289.6 µs CPU/req)	340 (142 MiB / 24.8% CPU · 85 MiB/s · 2915.5 µs CPU/req)	4,577 (135 MiB / 43.5% CPU · 1,144 MiB/s · 380.1 µs CPU/req)

Linux

Median of 3 repeats @ 482f5984. Source: Actions 37210533058.

Workload	TWP	nginx	HAProxy	Envoy	YARP
Keep-alive · tiny GET	23,748 (107 MiB / 49.8% CPU · 83.9 µs CPU/req)	🥇 27,308 (98 MiB / 42.0% CPU · 61.4 µs CPU/req)	26,617 (82 MiB / 44.8% CPU · 67.3 µs CPU/req)	17,004 (128 MiB / 59.3% CPU · 139.4 µs CPU/req)	20,349 (131 MiB / 50.0% CPU · 98.3 µs CPU/req)
New-connection · tiny GET	969 (124 MiB / 0.2% CPU · 10.3 µs CPU/req)	1,004 (98 MiB / 1.0% CPU · 39.1 µs CPU/req)	950 (85 MiB / 0.3% CPU · 13.1 µs CPU/req)	🥇 1,072 (128 MiB)	962 (147 MiB / 0.7% CPU · 30.4 µs CPU/req)
Keep-alive · 256 KiB GET	2,661 (132 MiB / 36.8% CPU · 665 MiB/s · 552.8 µs CPU/req)	1,701 (99 MiB / 53.3% CPU · 425 MiB/s · 1253.0 µs CPU/req)	🥇 2,926 (86 MiB / 32.5% CPU · 731 MiB/s · 444.9 µs CPU/req)	2,662 (144 MiB / 34.1% CPU · 665 MiB/s · 512.3 µs CPU/req)	2,122 (160 MiB / 45.4% CPU · 531 MiB/s · 855.0 µs CPU/req)

macOS

Median of 3 repeats on macos-15 @ 482f5984. Source: Actions 37210533058.

Workload	TWP	nginx	HAProxy	Envoy	YARP
Keep-alive · tiny GET	13,222 (123 MiB / 21.6% CPU · 48.9 µs CPU/req)	🥇 14,502 (84 MiB / 15.7% CPU · 32.5 µs CPU/req)	13,941 (108 MiB / 17.7% CPU · 38.1 µs CPU/req)	7,183 (121 MiB / 22.1% CPU · 92.5 µs CPU/req)	6,401 (160 MiB / 13.9% CPU · 65.3 µs CPU/req)
New-connection · tiny GET	125 (peak 130 · 107 MiB / 0.2% CPU · 38.5 µs CPU/req)	619 (86 MiB)	707 (105 MiB)	🥇 764 (131 MiB / 0.1% CPU · 2.4 µs CPU/req)	126 (158 MiB / 0.1% CPU · 11.9 µs CPU/req)
Keep-alive · 256 KiB GET	1,756 (217 MiB / 21.8% CPU · 439 MiB/s · 372.2 µs CPU/req)	1,898 (85 MiB / 18.2% CPU · 475 MiB/s · 287.6 µs CPU/req)	2,178 (116 MiB / 24.6% CPU · 544 MiB/s · 338.9 µs CPU/req)	🥇 2,258 (155 MiB / 26.4% CPU · 564 MiB/s · 350.4 µs CPU/req)	1,599 (181 MiB / 31.9% CPU · 400 MiB/s · 598.3 µs CPU/req)

Prefer TWP÷YARP in the table. Absolute RPS on GHA swings hard. New-connection is Darwin SslStream-bound for TWP and YARP (handshake p99 SLO 500 ms on macOS only — Win/Linux stay at 200 ms).

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 @ 482f5984 — 37210536200.

OS	Titanium	YARP	nginx	HAProxy	Envoy
Windows	🥇 42,967 (128 MiB / 31.4% CPU · 29.3 µs CPU/req)	27,337 (128 MiB / 47.0% CPU · 68.7 µs CPU/req)	0 (144 MiB / 24.8% CPU)	Not possible	Not possible
Linux	🥇 33,244 (168 MiB / 34.1% CPU · 41.1 µs CPU/req)	20,500 (152 MiB / 43.1% CPU · 84.1 µs CPU/req)	0 (110 MiB / 41.4% CPU)	15,252 (81 MiB / 44.6% CPU · 117.0 µs CPU/req)	16,458 (126 MiB / 56.0% CPU · 136.2 µs CPU/req)
macOS	0 (251 MiB / 10.8% CPU)	0 (263 MiB / 9.2% CPU)	0 (86 MiB / 2.5% CPU)	17,350 (123 MiB / 29.4% CPU · 50.9 µs CPU/req)	🥇 20,494 (121 MiB / 33.2% CPU · 48.6 µs CPU/req)

Unary gRPC (H2 TLS → h2c)

Unary Echo RPC/s @ c=64 over H2 TLS → h2c (edge TLS, cleartext H2 origin). Peers: Titanium, YARP, HAProxy, Envoy. Mode: compare-grpc (*-grpc-h2c). Median of 3 repeats @ 482f5984 — 37210536200.

Not possible: nginx column omitted (no H2 upstream).

OS	Titanium	YARP	HAProxy	Envoy
Windows	🥇 47,293 (129 MiB / 33.0% CPU · 27.9 µs CPU/req)	29,812 (136 MiB / 44.9% CPU · 60.2 µs CPU/req)	Not possible	Not possible
Linux	🥇 39,632 (163 MiB / 33.7% CPU · 34.0 µs CPU/req)	23,644 (153 MiB / 44.0% CPU · 74.5 µs CPU/req)	18,026 (83 MiB / 43.9% CPU · 97.3 µs CPU/req)	17,986 (125 MiB / 56.5% CPU · 125.7 µs CPU/req)
macOS	8,312 (273 MiB / 7.4% CPU · 26.7 µs CPU/req)	8,959 (235 MiB / 19.6% CPU · 65.6 µs CPU/req)	6,113 (110 MiB / 21.3% CPU · 104.3 µs CPU/req)	🥇 9,239 (120 MiB / 15.6% CPU · 50.7 µs CPU/req)

WebSocket (H1 TLS → H1 TLS)

WebSocket echo round-trips/sec over dual-TLS H1 (proxy_ssl style). Mode: compare-ws-h1tls (*-duplex-ws-h1tls). Median of 3 repeats @ 482f5984. Source: 37210538613.

OS	Titanium	YARP	nginx	HAProxy	Envoy
Windows	🥇 29,526 (99 MiB / 42.1% CPU · 57.1 µs CPU/req)	27,745 (93 MiB / 44.1% CPU · 63.5 µs CPU/req)	13,706 (142 MiB / 24.8% CPU · 72.2 µs CPU/req)	Not possible	Not possible
Linux	40,864 (137 MiB / 41.9% CPU · 41.0 µs CPU/req)	37,220 (135 MiB / 44.0% CPU · 47.2 µs CPU/req)	48,858 (104 MiB / 34.4% CPU · 28.2 µs CPU/req)	48,324 (84 MiB / 37.1% CPU · 30.7 µs CPU/req)	🥇 49,049 (127 MiB / 37.5% CPU · 30.6 µs CPU/req)
macOS	6,336 (147 MiB / 14.8% CPU · 70.1 µs CPU/req)	6,772 (149 MiB / 16.4% CPU · 72.8 µs CPU/req)	🥇 12,168 (91 MiB / 10.3% CPU · 25.4 µs CPU/req)	6,870 (107 MiB / 11.8% CPU · 51.3 µs CPU/req)	6,721 (123 MiB / 18.3% CPU · 81.5 µs CPU/req)

WebSocket (H2 TLS 8441 → H1)

WebSocket echo over RFC 8441 extended CONNECT (H2 TLS client → H1 plain origin). Mode: compare-ws-h2 (*-duplex-ws-h2).

The probe client sets NoDelay and writes each HTTP/2 frame as one TLS record. Median of 3 repeats @ 482f5984. Source: 37210541741.

Not possible: nginx column omitted (no RFC 8441 extended CONNECT).

OS	Titanium	YARP	HAProxy	Envoy
Windows	🥇 34,615 (152 MiB / 44.4% CPU · 51.3 µs CPU/req)	0 (160 MiB / 19.9% CPU)	Not possible	Not possible
Linux	🥇 41,579 (204 MiB / 49.5% CPU · 47.6 µs CPU/req)	0 (191 MiB / 43.2% CPU)	36,813 (85 MiB / 41.2% CPU · 44.7 µs CPU/req)	0 (123 MiB / 0.3% CPU)
macOS	6,839 (260 MiB / 21.9% CPU · 96.0 µs CPU/req)	0 (186 MiB / 10.5% CPU)	🥇 9,231 (109 MiB / 16.7% CPU · 54.3 µs CPU/req)	0 (116 MiB / 0.4% CPU)

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	Linux only; one job per wiki row (feature arm + its baseline)
Pre-wiki smoke	compare-product-smoke (Linux, repeats=1)	~30–60 min; required before full product
Release / wiki	compare-product reverse via RPS suite	one job per wiki row on Win/Linux/macOS; MITM arms (twp-mitm-*) on Linux only, same job as that row's reverse peers
Unary gRPC	compare-grpc	one job per row (H2 TLS and h2c) on Win/Linux/macOS
WebSocket dual-TLS	compare-ws-h1tls	one row, three OS
WebSocket RFC 8441	compare-ws-h2	one row, three OS (nginx N/A)
Heavier tables	compare-bodies, post, lossy, arch, tls-cost, compare-saturation	one suite run each, one job per row, on Win/Linux/macOS

See PERF-GATES.md.

Arm shards (comparison groups)

--arm-shard takes a comparison-group key (h1c-h1c, from --print-groups), not an i/n index, so a re-run measures the same row after a peer install flake. i/n remains for smoke modes. Paste unions rps-csv-<os>-shard-* (paste-compare-product-wiki.ps1 -RunIds …). Table “Source: Actions” may list several run URLs for one table — do not mix SHAs. A single-row leg allows 60 minutes of ramp and 75 minutes overall. Free GitHub accounts run 20 jobs at once, 5 of them macOS. Give a full suite a quiet window: do not push pull requests or start other macOS jobs while it runs. Re-run infrastructure failures only (gh run rerun --failed). A failed gate is a finding and is never re-rolled; publish the last attempt, never the best of several. The RPS tables on this page are @ 482f5984: product 37210516555, saturation 37210519205, bodies 37210521790, POST 37210524513, lossy 37210527573, arch 37210530448, TLS cost 37210533058, gRPC 37210536200, WebSocket H1 TLS 37210538613, WebSocket H2 37210541741, editions 37210543774. Reverse product rows, bodies, POST, lossy, arch, TLS cost, gRPC, both WebSocket modes, and saturation run on Windows, Linux, and macOS. TWP MITM arms run on Linux only. compare-editions (CLI and Plus) runs on Linux only. compare-spot on pull requests to beta and stable stays on all three OS. The historical v2 MITM note below the product tables stays @ 41f4adee because those run ids are from that fix, not this re-measure.

One wiki row = one GHA job’s Client×Origin cell set: TWP + YARP + nginx + HAProxy + Envoy (when the OS can run them) stay on the same VM. On Linux that job also runs TWP Lite and Full, so Lite÷Reverse stays a same-job ratio. Windows and macOS product jobs omit twp-mitm-*. Shards split rows, not individual proxies — so TWP÷YARP remains a same-job ratio. Do not compare absolute RPS across shards (different VMs).

Gate thresholds

• Reverse product signal: TWP ÷ closest peer ≥ 0.50 (the peer among YARP, nginx, HAProxy, and Envoy whose sustain is nearest). Edition CLI/Plus ratios floor at 0.50.
• MITM overhead: Lite÷Reverse ≥ 0.25 and Full÷Reverse ≥ 0.25 (median of 3 GHA runs @ c=64). Absolute RPS moves with runner heat; ratios are the claim.
• Do not enable the HTTP/2 MITM multi-origin relay pool (MaxOriginHttp2ConnectionsPerAuthority > 1 under interception): Tip A/B showed Lite err%~29 and RSS blow-up, which fails the Lite/Full÷Reverse floors. Multi-origin remains gate-off compressed-relay only.
• Editions: see PERF-GATES.md and validate-edition-gates.ps1.

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).