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-intel 4-core / 14 GB. 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-intel). Laptop numbers are never mixed into these tables.
Peers use equivalent TLS/ALPN and streaming-friendly settings on the same loopback shape. HAProxy and Envoy are Linux/macOS only — Not possible on Windows, so their columns are omitted there.
MITM (HTTPS decryption with forged certificates) is Titanium-only; peers cannot MITM. Those tables show Titanium MITM overhead versus its own reverse path on the same wires.
Tiny keep-alive GET (~56-byte JSON) is the industry RPS shape (same class as wrk / TechEmpower). It is also real for small JSON APIs and health checks.
Same-protocol H2↔H2 / H3↔H3 on that shape is Titanium’s best case: with interception off, Titanium copies frames instead of decoding and re-encoding headers (peers do a full HTTP decode). Medals there are not the typical reverse-proxy job.
Typical reverse is H1 TLS→H1 or H2→H1 (~1.1× YARP on tiny GET). With larger bodies, see Heavier reverse: at 64 KiB H2 TLS→H2 TLS, Titanium is behind YARP (~0.69–0.76×).

How to read the tables

Bold RPS is sustain (last concurrency that still met error/latency SLOs). When peak differs, it appears in the same cell as (peak N · …) with RSS/CPU. When sustain equals peak, peak is omitted.
🥇 = best among Titanium / nginx / YARP on Reverse rows (highest RPS; on a tie, lower memory then lower CPU%). Gold medals are per cell on tiny GET — an H2↔H2 gold is that frame-copy best case, not “Titanium is 1.7× on all reverse.” For larger-body H2→H2, see the heavier tables.
MITM tables are Titanium-only. Lite÷Reverse / Full÷Reverse = Titanium MITM sustain ÷ Titanium reverse sustain on the same Client×Origin pair — the overhead of decrypting and intercepting versus bare reverse, not versus nginx or YARP.
Not possible = cannot do that path. Not measured = path exists but no published number yet. When every row would be Not possible for a peer, that column is omitted and a note above the table explains why.

Why this comparison is fair
How to read the tables
Measurement environment
Windows — Titanium vs nginx vs YARP
Linux — Titanium vs nginx vs HAProxy vs Envoy vs YARP
macOS — Titanium vs nginx vs HAProxy vs Envoy vs YARP
Editions (CLI / Plus / Intercept)
Cross-version (7.0 vs 6.0)
Heavier reverse workloads
Unary gRPC (H2 TLS)
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 the 4-core-class public-repo GitHub-hosted runners: Windows / Linux at 4 vCPU / 16 GiB / 14 GB SSD, macOS at macos-15-intel 4-core / 14 GB (not macos-latest). Same harness knobs: warmup 2s / measure 8s; concurrency 8, 16, 32, 64; median of 3 repeats. Prefer Titanium÷YARP / Titanium÷nginx ratios over absolute RPS.

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

Windows (GitHub-hosted `windows-latest`)


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

Linux (GitHub-hosted `ubuntu-latest`)


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

macOS (GitHub-hosted `macos-15-intel`)


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

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

Saturation control

Calibration for the shared 4 vCPU loopback shape: how close client + origin are to saturated before ranking reverse peers. Tiny keep-alive GET. Median of 3 repeats @ 8bfa7852 — 35092854050. 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	81,217 _{(55 MiB / 45.3% CPU)}	100.0%
origin-direct-bombardier	bombardier	61,721 _{(56 MiB / 25.1% CPU)}	76.0%
bare-reverse-http1	dotnet-httpclient	39,529 _{(56 MiB / 44.1% CPU)}	48.7%
nginx-reverse-http1	dotnet-httpclient	24,677 _{(125 MiB / 24.9% CPU)}	30.4%
yarp-reverse-http1	dotnet-httpclient	34,602 _{(90 MiB / 49.2% CPU)}	42.6%
twp-reverse-http1	dotnet-httpclient	🥇 40,685 _{(76 MiB / 50.9% CPU)}	50.1%

Linux (ubuntu-latest)

Arm	Generator	RPS	% of origin-HttpClient
origin-direct	dotnet-httpclient	131,605 _{(81 MiB / 45.0% CPU)}	100.0%
origin-direct-bombardier	bombardier	81,228 _{(80 MiB / 38.3% CPU)}	61.7%
bare-reverse-http1	dotnet-httpclient	58,771 _{(70 MiB / 46.0% CPU)}	44.7%
nginx-reverse-http1	dotnet-httpclient	🥇 74,446 _{(77 MiB / 39.5% CPU)}	56.6%
yarp-reverse-http1	dotnet-httpclient	55,568 _{(115 MiB / 48.0% CPU)}	42.2%
twp-reverse-http1	dotnet-httpclient	63,570 _{(92 MiB / 50.3% CPU)}	48.3%

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

Block B — H2 TLS→H1

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

Windows (windows-latest)

Arm	Generator	RPS	÷YARP	÷nginx
nginx-reverse-http2	dotnet-httpclient	14,697 _{(141 MiB / 24.4% CPU)}	0.32×	1.00×
yarp-reverse-http2	dotnet-httpclient	45,876 _{(96 MiB / 50.8% CPU)}	1.00×	3.12×
twp-reverse-http2-cleartext	dotnet-httpclient	🥇 51,822 _{(103 MiB / 52.9% CPU)}	1.13×	3.53×

Linux (ubuntu-latest)

Arm	Generator	RPS	÷YARP	÷nginx
nginx-reverse-http2	dotnet-httpclient	29,423 _{(102 MiB / 18.9% CPU)}	0.46×	1.00×
yarp-reverse-http2	dotnet-httpclient	64,174 _{(123 MiB / 46.6% CPU)}	1.00×	2.18×
twp-reverse-http2-cleartext	dotnet-httpclient	🥇 66,465 _{(123 MiB / 51.4% CPU)}	1.04×	2.26×

Block C — H3→H1

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

Windows (windows-latest)

Arm	Generator	RPS	÷YARP	÷nginx
nginx-reverse-http3-cleartext	dotnet-httpclient	Not possible (no QUIC)	—	—
yarp-reverse-http3-cleartext	dotnet-httpclient	24,601 _{(148 MiB / 50.5% CPU)}	1.00×	—
twp-reverse-http3-cleartext	dotnet-httpclient	🥇 24,701 _{(117 MiB / 43.9% CPU)}	1.00×	—

Linux (ubuntu-latest)

Arm	Generator	RPS	÷YARP	÷nginx
nginx-reverse-http3-cleartext	dotnet-httpclient	0 _{(peak 38,114 · 113 MiB / 20.9% CPU)}	0.96×	1.00×
yarp-reverse-http3-cleartext	dotnet-httpclient	39,807 _{(215 MiB / 48.7% CPU)}	1.00×	1.04×
twp-reverse-http3-cleartext	dotnet-httpclient	🥇 44,892 _{(177 MiB / 53.4% CPU)}	1.13×	1.18×

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 @ 8bfa7852 — compare-product 35092827644. 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  (MiB / CPU%). 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	🥇 31,552 _{(75 MiB / 51.1% CPU)}	19,301 _{(125 MiB / 25% CPU)}	26,979 _{(90 MiB / 48.1% CPU)}
HTTP/1 · plain	HTTP/1 · TLS	🥇 20,447 _{(90 MiB / 52.6% CPU)}	8,419 _{(136 MiB / 24.6% CPU)}	18,937 _{(100 MiB / 48.3% CPU)}
HTTP/1 · plain	HTTP/2 · plain	🥇 36,987 _{(108 MiB / 45.2% CPU)}	Not possible (no H2 upstream)	32,831 _{(93 MiB / 46.7% CPU)}
HTTP/1 · plain	HTTP/2 · TLS	🥇 39,580 _{(117 MiB / 47.3% CPU)}	Not possible (no H2 upstream)	36,173 _{(98 MiB / 48.9% CPU)}
HTTP/1 · plain	HTTP/3 · QUIC	11,519 _{(104 MiB / 35% CPU)}	Not possible (no H3 upstream)	🥇 18,177 _{(118 MiB / 49.9% CPU)}
HTTP/1 · TLS	HTTP/1 · plain	🥇 20,182 _{(86 MiB / 53.9% CPU)}	8,534 _{(142 MiB / 24.7% CPU)}	18,049 _{(102 MiB / 49.4% CPU)}
HTTP/1 · TLS	HTTP/1 · TLS	🥇 23,518 _{(84 MiB / 47.8% CPU)}	9,376 _{(144 MiB / 24.8% CPU)}	20,816 _{(102 MiB / 51.2% CPU)}
HTTP/1 · TLS	HTTP/2 · plain	🥇 27,742 _{(115 MiB / 46% CPU)}	Not possible (no H2 upstream)	26,111 _{(108 MiB / 49.6% CPU)}
HTTP/1 · TLS	HTTP/2 · TLS	🥇 25,046 _{(119 MiB / 47% CPU)}	Not possible (no H2 upstream)	24,356 _{(108 MiB / 47.4% CPU)}
HTTP/1 · TLS	HTTP/3 · QUIC	🥇 19,385 _{(111 MiB / 50.9% CPU)}	Not possible (no H3 upstream)	18,399 _{(129 MiB / 49.7% CPU)}
HTTP/2 · plain	HTTP/1 · plain	🥇 34,785 _{(95 MiB / 53.6% CPU)}	9,225 _{(127 MiB / 24.8% CPU)}	31,535 _{(84 MiB / 53.4% CPU)}
HTTP/2 · plain	HTTP/1 · TLS	🥇 29,745 _{(98 MiB / 49% CPU)}	6,559 _{(138 MiB / 24.9% CPU)}	27,120 _{(96 MiB / 54% CPU)}
HTTP/2 · plain	HTTP/2 · plain	🥇 114,586 _{(59 MiB / 27.4% CPU)}	Not possible (no H2 upstream)	69,424 _{(94 MiB / 48.1% CPU)}
HTTP/2 · plain	HTTP/2 · TLS	🥇 86,454 _{(65 MiB / 28.5% CPU)}	Not possible (no H2 upstream)	55,488 _{(99 MiB / 46.8% CPU)}
HTTP/2 · plain	HTTP/3 · QUIC	🥇 31,645 _{(121 MiB / 52.4% CPU)}	Not possible (no H3 upstream)	28,884 _{(120 MiB / 52.8% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	🥇 39,258 _{(103 MiB / 51.6% CPU)}	11,247 _{(141 MiB / 24.3% CPU)}	35,070 _{(94 MiB / 49% CPU)}
HTTP/2 · TLS	HTTP/1 · TLS	🥇 29,467 _{(106 MiB / 54% CPU)}	6,379 _{(145 MiB / 24.6% CPU)}	25,223 _{(97 MiB / 51.8% CPU)}
HTTP/2 · TLS	HTTP/2 · plain	🥇 101,094 _{(78 MiB / 26.5% CPU)}	Not possible (no H2 upstream)	54,554 _{(99 MiB / 50.8% CPU)}
HTTP/2 · TLS	HTTP/2 · TLS	🥇 88,782 _{(75 MiB / 28.5% CPU)}	Not possible (no H2 upstream)	53,218 _{(103 MiB / 48.9% CPU)}
HTTP/2 · TLS	HTTP/3 · QUIC	🥇 30,154 _{(129 MiB / 54.2% CPU)}	Not possible (no H3 upstream)	26,201 _{(123 MiB / 51.9% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	🥇 14,399 _{(101 MiB / 45.8% CPU)}	Not possible (no QUIC)	14,330 _{(142 MiB / 51.1% CPU)}
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 17,592 _{(116 MiB / 44.1% CPU)}	Not possible (no QUIC)	15,607 _{(147 MiB / 48.8% CPU)}
HTTP/3 · QUIC	HTTP/2 · plain	🥇 34,116 _{(136 MiB / 47.7% CPU)}	Not possible (no H2 upstream)	23,176 _{(151 MiB / 47.4% CPU)}
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 31,135 _{(137 MiB / 47.1% CPU)}	Not possible (no H2 upstream)	21,466 _{(150 MiB / 48.8% CPU)}
HTTP/3 · QUIC	HTTP/3 · QUIC	17,411 _{(119 MiB / 41.5% CPU)}	Not possible (no H3 upstream)	🥇 18,619 _{(163 MiB / 48.3% CPU)}

MITM (TWP only)

Same Client×Origin wires with interception on (compare-product 35092827644). Lite = no-op handlers (unchanged-lite finish). Full = append-only header mutation (harness: one probe header each way; product: generic append-only relay via MitmCompressedRelayHelper). nginx/HAProxy/Envoy/YARP cannot MITM. Lite÷Reverse / Full÷Reverse vs bare reverse (same job / comparison-group shard). Completion gate: Lite ≥ 0.50× and Full ≥ 0.50× reverse sustain @ c=64 (median of 3 GHA runs); reverse TWP÷YARP ≥ 0.75× (no terminate-peer gate).

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

v2 drop-only + non-unique append (2026-08-27): MitmStaticRebuildHelper rebuilds static HPACK/QPACK after 1–4 unique header drops; trailing non-unique appends stay on compressed relay. @ 8bfa7852: 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	30,988 _{(80 MiB / 46.4% CPU)}	30,492 _{(82 MiB / 48.2% CPU)}	0.98×	0.97×
HTTP/1 · plain	HTTP/1 · TLS	20,384 _{(95 MiB / 52.8% CPU)}	20,198 _{(95 MiB / 52.5% CPU)}	1×	0.99×
HTTP/1 · plain	HTTP/2 · plain	35,601 _{(110 MiB / 48% CPU)}	35,122 _{(105 MiB / 49.1% CPU)}	0.96×	0.95×
HTTP/1 · plain	HTTP/2 · TLS	39,383 _{(120 MiB / 48.3% CPU)}	38,670 _{(120 MiB / 47.9% CPU)}	1×	0.98×
HTTP/1 · plain	HTTP/3 · QUIC	18,054 _{(103 MiB / 53.2% CPU)}	17,685 _{(107 MiB / 52.2% CPU)}	1.57×	1.54×
HTTP/1 · TLS	HTTP/1 · plain	19,722 _{(94 MiB / 51.6% CPU)}	19,661 _{(96 MiB / 51.5% CPU)}	0.98×	0.97×
HTTP/1 · TLS	HTTP/1 · TLS	23,590 _{(95 MiB / 48.8% CPU)}	23,182 _{(96 MiB / 46.8% CPU)}	1×	0.99×
HTTP/1 · TLS	HTTP/2 · plain	27,300 _{(110 MiB / 48.4% CPU)}	26,872 _{(112 MiB / 47.3% CPU)}	0.98×	0.97×
HTTP/1 · TLS	HTTP/2 · TLS	25,107 _{(124 MiB / 45.4% CPU)}	24,748 _{(127 MiB / 47.5% CPU)}	1×	0.99×
HTTP/1 · TLS	HTTP/3 · QUIC	18,795 _{(112 MiB / 50.6% CPU)}	18,674 _{(115 MiB / 53.7% CPU)}	0.97×	0.96×
HTTP/2 · plain	HTTP/1 · plain	33,811 _{(97 MiB / 55.8% CPU)}	33,355 _{(100 MiB / 55.9% CPU)}	0.97×	0.96×
HTTP/2 · plain	HTTP/1 · TLS	29,166 _{(105 MiB / 53.2% CPU)}	28,519 _{(103 MiB / 55.1% CPU)}	0.98×	0.96×
HTTP/2 · plain	HTTP/2 · plain	90,897 _{(69 MiB / 41.3% CPU)}	84,998 _{(70 MiB / 43% CPU)}	0.79×	0.74×
HTTP/2 · plain	HTTP/2 · TLS	70,465 _{(78 MiB / 38.9% CPU)}	66,829 _{(79 MiB / 38.1% CPU)}	0.82×	0.77×
HTTP/2 · plain	HTTP/3 · QUIC	31,297 _{(116 MiB / 54.3% CPU)}	30,424 _{(115 MiB / 55.6% CPU)}	0.99×	0.96×
HTTP/2 · TLS	HTTP/1 · plain	38,437 _{(103 MiB / 53.4% CPU)}	37,561 _{(112 MiB / 51.9% CPU)}	0.98×	0.96×
HTTP/2 · TLS	HTTP/1 · TLS	28,202 _{(105 MiB / 55.9% CPU)}	27,538 _{(102 MiB / 54.8% CPU)}	0.96×	0.93×
HTTP/2 · TLS	HTTP/2 · plain	79,360 _{(87 MiB / 40.9% CPU)}	74,807 _{(87 MiB / 40.4% CPU)}	0.79×	0.74×
HTTP/2 · TLS	HTTP/2 · TLS	74,209 _{(81 MiB / 39.1% CPU)}	70,572 _{(85 MiB / 39.3% CPU)}	0.84×	0.79×
HTTP/2 · TLS	HTTP/3 · QUIC	30,269 _{(131 MiB / 53.9% CPU)}	29,544 _{(130 MiB / 52.9% CPU)}	1×	0.98×
HTTP/3 · QUIC	HTTP/1 · plain	13,942 _{(113 MiB / 45.5% CPU)}	13,712 _{(111 MiB / 46.6% CPU)}	0.97×	0.95×
HTTP/3 · QUIC	HTTP/1 · TLS	16,217 _{(120 MiB / 44% CPU)}	15,113 _{(124 MiB / 44.8% CPU)}	0.92×	0.86×
HTTP/3 · QUIC	HTTP/2 · plain	31,662 _{(139 MiB / 51.8% CPU)}	29,770 _{(139 MiB / 50.7% CPU)}	0.93×	0.87×
HTTP/3 · QUIC	HTTP/2 · TLS	28,483 _{(130 MiB / 49.9% CPU)}	27,720 _{(133 MiB / 48.2% CPU)}	0.91×	0.89×
HTTP/3 · QUIC	HTTP/3 · QUIC	17,584 _{(123 MiB / 45.2% CPU)}	23,082 _{(123 MiB / 46.4% CPU)}	1.01×	1.33×

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 @ 8bfa7852 — compare-product 35092827644. 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	52,118 _{(95 MiB / 48.8% CPU)}	🥇 70,158 _{(76 MiB / 38.3% CPU)}	66,062 _{(64 MiB / 41.8% CPU)}	38,126 _{(116 MiB / 59.3% CPU)}	47,079 _{(115 MiB / 49.4% CPU)}
HTTP/1 · plain	HTTP/1 · TLS	23,523 _{(107 MiB / 51.1% CPU)}	28,019 _{(92 MiB / 42.1% CPU)}	🥇 28,193 _{(68 MiB / 43.6% CPU)}	17,450 _{(118 MiB / 59% CPU)}	21,545 _{(127 MiB / 50.3% CPU)}
HTTP/1 · plain	HTTP/2 · plain	🥇 40,142 _{(129 MiB / 51.1% CPU)}	Not possible (no H2 upstream)	26,857 _{(65 MiB / 42.4% CPU)}	22,067 _{(116 MiB / 63.3% CPU)}	35,268 _{(126 MiB / 48.8% CPU)}
HTTP/1 · plain	HTTP/2 · TLS	51,676 _{(158 MiB / 48.4% CPU)}	Not possible (no H2 upstream)	🥇 54,114 _{(68 MiB / 41% CPU)}	37,450 _{(117 MiB / 57.4% CPU)}	48,422 _{(131 MiB / 48.2% CPU)}
HTTP/1 · plain	HTTP/3 · QUIC	🥇 23,529 _{(143 MiB / 53.6% CPU)}	Not possible (no H3 upstream)	Not possible (no H3 upstream)	10,746 _{(120 MiB / 56.1% CPU)}	21,315 _{(156 MiB / 49.2% CPU)}
HTTP/1 · TLS	HTTP/1 · plain	22,984 _{(112 MiB / 49.8% CPU)}	🥇 27,706 _{(100 MiB / 41.6% CPU)}	27,584 _{(83 MiB / 43.2% CPU)}	17,122 _{(127 MiB / 58% CPU)}	20,033 _{(134 MiB / 50.2% CPU)}
HTTP/1 · TLS	HTTP/1 · TLS	35,195 _{(112 MiB / 45.8% CPU)}	43,625 _{(105 MiB / 38.9% CPU)}	🥇 44,407 _{(84 MiB / 40.7% CPU)}	30,389 _{(127 MiB / 53.8% CPU)}	31,410 _{(143 MiB / 48.3% CPU)}
HTTP/1 · TLS	HTTP/2 · plain	🥇 27,813 _{(141 MiB / 50.6% CPU)}	Not possible (no H2 upstream)	21,585 _{(84 MiB / 42.3% CPU)}	18,519 _{(126 MiB / 58.2% CPU)}	25,000 _{(148 MiB / 49.7% CPU)}
HTTP/1 · TLS	HTTP/2 · TLS	🥇 24,173 _{(155 MiB / 48.7% CPU)}	Not possible (no H2 upstream)	23,923 _{(84 MiB / 43.3% CPU)}	17,323 _{(126 MiB / 57.5% CPU)}	22,008 _{(148 MiB / 47.6% CPU)}
HTTP/1 · TLS	HTTP/3 · QUIC	🥇 26,290 _{(157 MiB / 51.9% CPU)}	Not possible (no H3 upstream)	Not possible (no H3 upstream)	17,802 _{(130 MiB / 52.4% CPU)}	24,262 _{(171 MiB / 49.8% CPU)}
HTTP/2 · plain	HTTP/1 · plain	🥇 36,694 _{(120 MiB / 53.3% CPU)}	16,440 _{(79 MiB / 19% CPU)}	23,041 _{(66 MiB / 24.6% CPU)}	13,601 _{(118 MiB / 23.2% CPU)}	34,322 _{(114 MiB / 50.2% CPU)}
HTTP/2 · plain	HTTP/1 · TLS	🥇 27,336 _{(129 MiB / 51% CPU)}	12,859 _{(100 MiB / 19.6% CPU)}	18,613 _{(70 MiB / 24.6% CPU)}	12,985 _{(119 MiB / 23.4% CPU)}	25,795 _{(127 MiB / 51% CPU)}
HTTP/2 · plain	HTTP/2 · plain	🥇 118,021 _{(89 MiB / 36% CPU)}	Not possible (no H2 upstream)	48,169 _{(68 MiB / 24.2% CPU)}	25,730 _{(116 MiB / 21.3% CPU)}	68,950 _{(137 MiB / 48% CPU)}
HTTP/2 · plain	HTTP/2 · TLS	🥇 58,227 _{(86 MiB / 35.1% CPU)}	Not possible (no H2 upstream)	19,761 _{(69 MiB / 24.6% CPU)}	15,929 _{(118 MiB / 21.6% CPU)}	39,519 _{(128 MiB / 45.6% CPU)}
HTTP/2 · plain	HTTP/3 · QUIC	🥇 27,742 _{(147 MiB / 50.5% CPU)}	Not possible (no H3 upstream)	Not possible (no H3 upstream)	4,280 _{(121 MiB / 24.9% CPU)}	26,149 _{(153 MiB / 46.4% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	🥇 49,222 _{(125 MiB / 49.5% CPU)}	30,623 _{(102 MiB / 19.1% CPU)}	40,300 _{(81 MiB / 24.3% CPU)}	22,680 _{(129 MiB / 22.7% CPU)}	45,846 _{(123 MiB / 47.7% CPU)}
HTTP/2 · TLS	HTTP/1 · TLS	🥇 26,240 _{(122 MiB / 50.9% CPU)}	12,497 _{(110 MiB / 19.9% CPU)}	16,768 _{(85 MiB / 24.4% CPU)}	13,048 _{(128 MiB / 23.3% CPU)}	23,132 _{(133 MiB / 50.3% CPU)}
HTTP/2 · TLS	HTTP/2 · plain	🥇 75,608 _{(99 MiB / 36.6% CPU)}	Not possible (no H2 upstream)	21,385 _{(82 MiB / 24.4% CPU)}	17,016 _{(126 MiB / 22.1% CPU)}	39,430 _{(130 MiB / 47.6% CPU)}
HTTP/2 · TLS	HTTP/2 · TLS	🥇 81,336 _{(101 MiB / 34.1% CPU)}	Not possible (no H2 upstream)	38,308 _{(82 MiB / 24.2% CPU)}	24,040 _{(126 MiB / 21.6% CPU)}	46,396 _{(129 MiB / 44% CPU)}
HTTP/2 · TLS	HTTP/3 · QUIC	🥇 25,960 _{(154 MiB / 50.3% CPU)}	Not possible (no H3 upstream)	Not possible (no H3 upstream)	4,339 _{(129 MiB / 24.8% CPU)}	22,717 _{(154 MiB / 47% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	20,471 _{(147 MiB / 50% CPU)}	0 _{(peak 15,362 · 106 MiB / 22.5% CPU)}	🥇 23,011 _{(87 MiB / 26.3% CPU)}	3,989 _{(136 MiB / 24.8% CPU)}	18,716 _{(183 MiB / 50.2% CPU)}
HTTP/3 · QUIC	HTTP/1 · TLS	23,817 _{(165 MiB / 45.2% CPU)}	0 _{(peak 27,931 · 123 MiB / 20.1% CPU)}	🥇 29,255 _{(91 MiB / 25.8% CPU)}	14 _{(136 MiB / 0.1% CPU)}	22,334 _{(203 MiB / 47.7% CPU)}
HTTP/3 · QUIC	HTTP/2 · plain	🥇 28,969 _{(158 MiB / 53% CPU)}	Not possible (no H2 upstream)	24,863 _{(86 MiB / 25.8% CPU)}	26 _{(130 MiB / 0.2% CPU)}	23,885 _{(197 MiB / 48.2% CPU)}
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 25,108 _{(157 MiB / 50.6% CPU)}	Not possible (no H2 upstream)	20,697 _{(88 MiB / 25.4% CPU)}	1,471 _{(132 MiB / 7.8% CPU)}	21,179 _{(196 MiB / 47.3% CPU)}
HTTP/3 · QUIC	HTTP/3 · QUIC	🥇 22,764 _{(170 MiB / 44.9% CPU)}	Not possible (no H3 upstream)	Not possible (no H3 upstream)	448 _{(130 MiB / 2.2% CPU)}	19,268 _{(210 MiB / 47.5% CPU)}

MITM (TWP only)

Same Client×Origin wires with interception on (compare-product 35092827644). Lite = no-op handlers (unchanged-lite finish). Full = append-only header mutation (harness: one probe header each way; product: generic append-only relay via MitmCompressedRelayHelper). nginx/HAProxy/Envoy/YARP cannot MITM. Lite÷Reverse / Full÷Reverse vs bare reverse (same job / comparison-group shard). Completion gate: Lite ≥ 0.50× and Full ≥ 0.50× reverse sustain @ c=64 (median of 3 GHA runs); reverse TWP÷YARP ≥ 0.75× (no terminate-peer gate).

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

v2 drop-only + non-unique append (2026-08-27): MitmStaticRebuildHelper rebuilds static HPACK/QPACK after 1–4 unique header drops; trailing non-unique appends stay on compressed relay. @ 8bfa7852: 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	51,441 _{(97 MiB / 50.2% CPU)}	50,516 _{(98 MiB / 49.8% CPU)}	0.99×	0.97×
HTTP/1 · plain	HTTP/1 · TLS	23,548 _{(112 MiB / 52.2% CPU)}	23,482 _{(113 MiB / 51.3% CPU)}	1×	1×
HTTP/1 · plain	HTTP/2 · plain	40,450 _{(132 MiB / 53.8% CPU)}	38,305 _{(126 MiB / 53% CPU)}	1.01×	0.95×
HTTP/1 · plain	HTTP/2 · TLS	50,652 _{(153 MiB / 50.5% CPU)}	49,728 _{(154 MiB / 50.8% CPU)}	0.98×	0.96×
HTTP/1 · plain	HTTP/3 · QUIC	22,774 _{(142 MiB / 54.2% CPU)}	22,822 _{(137 MiB / 54.3% CPU)}	0.97×	0.97×
HTTP/1 · TLS	HTTP/1 · plain	23,114 _{(115 MiB / 50.9% CPU)}	22,808 _{(116 MiB / 50.3% CPU)}	1.01×	0.99×
HTTP/1 · TLS	HTTP/1 · TLS	34,966 _{(117 MiB / 46.7% CPU)}	34,270 _{(118 MiB / 46.8% CPU)}	0.99×	0.97×
HTTP/1 · TLS	HTTP/2 · plain	27,488 _{(143 MiB / 51.3% CPU)}	26,553 _{(148 MiB / 51.2% CPU)}	0.99×	0.95×
HTTP/1 · TLS	HTTP/2 · TLS	23,559 _{(170 MiB / 48.9% CPU)}	23,089 _{(169 MiB / 48.7% CPU)}	0.97×	0.96×
HTTP/1 · TLS	HTTP/3 · QUIC	25,937 _{(156 MiB / 52.5% CPU)}	26,697 _{(161 MiB / 53.2% CPU)}	0.99×	1.02×
HTTP/2 · plain	HTTP/1 · plain	36,272 _{(120 MiB / 54.6% CPU)}	34,343 _{(120 MiB / 54.4% CPU)}	0.99×	0.94×
HTTP/2 · plain	HTTP/1 · TLS	27,164 _{(124 MiB / 52.4% CPU)}	26,445 _{(129 MiB / 52.2% CPU)}	0.99×	0.97×
HTTP/2 · plain	HTTP/2 · plain	88,034 _{(101 MiB / 40.5% CPU)}	81,934 _{(98 MiB / 40.2% CPU)}	0.75×	0.69×
HTTP/2 · plain	HTTP/2 · TLS	48,275 _{(95 MiB / 40.2% CPU)}	45,097 _{(98 MiB / 40.2% CPU)}	0.83×	0.77×
HTTP/2 · plain	HTTP/3 · QUIC	27,595 _{(143 MiB / 50.5% CPU)}	27,350 _{(147 MiB / 51.1% CPU)}	0.99×	0.99×
HTTP/2 · TLS	HTTP/1 · plain	48,535 _{(132 MiB / 50.6% CPU)}	46,791 _{(134 MiB / 50% CPU)}	0.99×	0.95×
HTTP/2 · TLS	HTTP/1 · TLS	25,779 _{(123 MiB / 51.8% CPU)}	25,125 _{(129 MiB / 52.1% CPU)}	0.98×	0.96×
HTTP/2 · TLS	HTTP/2 · plain	55,993 _{(110 MiB / 41.3% CPU)}	53,218 _{(114 MiB / 41% CPU)}	0.74×	0.7×
HTTP/2 · TLS	HTTP/2 · TLS	64,140 _{(113 MiB / 36.2% CPU)}	60,738 _{(112 MiB / 36.2% CPU)}	0.79×	0.75×
HTTP/2 · TLS	HTTP/3 · QUIC	26,305 _{(160 MiB / 51.5% CPU)}	25,709 _{(152 MiB / 50.6% CPU)}	1.01×	0.99×
HTTP/3 · QUIC	HTTP/1 · plain	19,415 _{(144 MiB / 50.9% CPU)}	19,586 _{(145 MiB / 50.1% CPU)}	0.95×	0.96×
HTTP/3 · QUIC	HTTP/1 · TLS	23,253 _{(171 MiB / 47.3% CPU)}	22,767 _{(170 MiB / 46% CPU)}	0.98×	0.96×
HTTP/3 · QUIC	HTTP/2 · plain	27,212 _{(160 MiB / 54.2% CPU)}	26,934 _{(163 MiB / 54.5% CPU)}	0.94×	0.93×
HTTP/3 · QUIC	HTTP/2 · TLS	23,551 _{(152 MiB / 51.9% CPU)}	22,624 _{(153 MiB / 51.9% CPU)}	0.94×	0.9×
HTTP/3 · QUIC	HTTP/3 · QUIC	21,049 _{(171 MiB / 48.3% CPU)}	20,830 _{(171 MiB / 48.2% CPU)}	0.92×	0.92×

macOS — Titanium vs nginx vs HAProxy vs Envoy vs YARP

Reverse

Median of 3 repeats on macos-15-intel (4-core / 14 GB). Bare reverse 5×5 @ 8bfa7852 — compare-product 35092827644. 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  (MiB / CPU%). The RPS workflow installs Homebrew nginx (http_v3_module), Homebrew HAProxy with USE_QUIC (3.2 source fallback), Envoy (Homebrew bottle or pinned darwin-amd64 1.36.7), Homebrew libmsquic (+ DYLD_*), and YARP. Do not publish from macos-latest (3-core / 7 GB). Product 5×5 is ~56-byte JSON keep-alive GET; H2/H3 same-protocol cells are mostly header work with a tiny body (Titanium best case) — see Why this comparison is fair.

Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
HTTP/1 · plain	HTTP/1 · plain	🥇 16,348 _{(84 MiB / 39.4% CPU)}	9,898 _{(52 MiB / 18.6% CPU)}	15,804 _{(50 MiB / 31.4% CPU)}	6,117 _{(72 MiB / 41.8% CPU)}	13,286 _{(104 MiB / 41.2% CPU)}
HTTP/1 · plain	HTTP/1 · TLS	🥇 12,783 _{(94 MiB / 43.1% CPU)}	5,967 _{(74 MiB / 18.6% CPU)}	12,525 _{(55 MiB / 33.2% CPU)}	5,670 _{(73 MiB / 48% CPU)}	9,909 _{(124 MiB / 40.9% CPU)}
HTTP/1 · plain	HTTP/2 · plain	🥇 20,904 _{(88 MiB / 43.3% CPU)}	Not possible (no H2 upstream)	8,778 _{(51 MiB / 30.6% CPU)}	6,122 _{(71 MiB / 44.3% CPU)}	15,420 _{(112 MiB / 40.6% CPU)}
HTTP/1 · plain	HTTP/2 · TLS	13,267 _{(101 MiB / 37.9% CPU)}	Not possible (no H2 upstream)	12,220 _{(52 MiB / 31.2% CPU)}	7,310 _{(73 MiB / 48.7% CPU)}	🥇 18,474 _{(116 MiB / 41.3% CPU)}
HTTP/1 · plain	HTTP/3 · QUIC	4,693 _{(89 MiB / 48.9% CPU)}	Not possible (no H3 upstream)	Not possible (no H3 upstream)	3,091 _{(75 MiB / 23.7% CPU)}	🥇 5,445 _{(115 MiB / 38.5% CPU)}
HTTP/1 · TLS	HTTP/1 · plain	7,955 _{(94 MiB / 34.9% CPU)}	6,371 _{(74 MiB / 25.2% CPU)}	7,738 _{(65 MiB / 31.1% CPU)}	6,444 _{(80 MiB / 41.2% CPU)}	🥇 8,665 _{(132 MiB / 38.8% CPU)}
HTTP/1 · TLS	HTTP/1 · TLS	8,977 _{(98 MiB / 36.1% CPU)}	5,494 _{(83 MiB / 18.9% CPU)}	🥇 9,125 _{(68 MiB / 35.1% CPU)}	3,695 _{(81 MiB / 37.8% CPU)}	7,820 _{(139 MiB / 37.4% CPU)}
HTTP/1 · TLS	HTTP/2 · plain	9,860 _{(97 MiB / 39.1% CPU)}	Not possible (no H2 upstream)	7,800 _{(66 MiB / 32.1% CPU)}	5,521 _{(79 MiB / 45.2% CPU)}	🥇 10,651 _{(118 MiB / 34.5% CPU)}
HTTP/1 · TLS	HTTP/2 · TLS	7,860 _{(163 MiB / 37% CPU)}	Not possible (no H2 upstream)	7,448 _{(66 MiB / 35.7% CPU)}	4,624 _{(79 MiB / 40.9% CPU)}	🥇 11,054 _{(127 MiB / 34.9% CPU)}
HTTP/1 · TLS	HTTP/3 · QUIC	2,705 _{(109 MiB / 46.2% CPU)}	Not possible (no H3 upstream)	Not possible (no H3 upstream)	3,636 _{(81 MiB / 32.3% CPU)}	🥇 4,619 _{(123 MiB / 35.2% CPU)}
HTTP/2 · plain	HTTP/1 · plain	🥇 16,372 _{(91 MiB / 42.4% CPU)}	15,429 _{(58 MiB / 18.7% CPU)}	10,220 _{(56 MiB / 20.3% CPU)}	3,692 _{(75 MiB / 22.4% CPU)}	14,817 _{(105 MiB / 44.2% CPU)}
HTTP/2 · plain	HTTP/1 · TLS	🥇 12,382 _{(94 MiB / 46.4% CPU)}	7,196 _{(85 MiB / 18.3% CPU)}	5,558 _{(59 MiB / 20.8% CPU)}	3,155 _{(77 MiB / 21.5% CPU)}	12,044 _{(125 MiB / 45.8% CPU)}
HTTP/2 · plain	HTTP/2 · plain	🥇 36,700 _{(73 MiB / 30.8% CPU)}	Not possible (no H2 upstream)	9,528 _{(57 MiB / 19.9% CPU)}	6,036 _{(72 MiB / 22% CPU)}	24,939 _{(111 MiB / 40.4% CPU)}
HTTP/2 · plain	HTTP/2 · TLS	🥇 30,656 _{(77 MiB / 30.9% CPU)}	Not possible (no H2 upstream)	8,666 _{(58 MiB / 20.9% CPU)}	5,469 _{(73 MiB / 21.3% CPU)}	26,491 _{(114 MiB / 38.9% CPU)}
HTTP/2 · plain	HTTP/3 · QUIC	5,092 _{(95 MiB / 56.1% CPU)}	Not possible (no H3 upstream)	Not possible (no H3 upstream)	1,973 _{(75 MiB / 22.8% CPU)}	🥇 5,453 _{(118 MiB / 39.1% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	🥇 15,856 _{(92 MiB / 44.7% CPU)}	9,770 _{(73 MiB / 18.1% CPU)}	8,427 _{(67 MiB / 21.2% CPU)}	3,491 _{(81 MiB / 21.4% CPU)}	12,707 _{(115 MiB / 43.2% CPU)}
HTTP/2 · TLS	HTTP/1 · TLS	🥇 13,319 _{(96 MiB / 44.8% CPU)}	6,541 _{(91 MiB / 19% CPU)}	8,793 _{(69 MiB / 21.4% CPU)}	3,229 _{(83 MiB / 21.7% CPU)}	11,692 _{(123 MiB / 44.6% CPU)}
HTTP/2 · TLS	HTTP/2 · plain	🥇 27,511 _{(84 MiB / 31.1% CPU)}	Not possible (no H2 upstream)	8,259 _{(68 MiB / 20.9% CPU)}	5,966 _{(79 MiB / 21.7% CPU)}	19,566 _{(112 MiB / 42.1% CPU)}
HTTP/2 · TLS	HTTP/2 · TLS	🥇 30,544 _{(84 MiB / 31.1% CPU)}	Not possible (no H2 upstream)	7,040 _{(68 MiB / 21.3% CPU)}	5,660 _{(79 MiB / 21.4% CPU)}	19,845 _{(116 MiB / 40.9% CPU)}
HTTP/2 · TLS	HTTP/3 · QUIC	5,713 _{(106 MiB / 45.9% CPU)}	Not possible (no H3 upstream)	Not possible (no H3 upstream)	2,513 _{(80 MiB / 22.9% CPU)}	🥇 7,097 _{(121 MiB / 40.4% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	4,598 _{(100 MiB / 46.2% CPU)}	0 _{(peak 6,020 · 62 MiB / 13.4% CPU)}	🥇 9,237 _{(67 MiB / 23.9% CPU)}	1,238 _{(86 MiB / 23.4% CPU)}	4,568 _{(183 MiB / 39% CPU)}
HTTP/3 · QUIC	HTTP/1 · TLS	4,263 _{(117 MiB / 43.4% CPU)}	0 _{(peak 7,101 · 67 MiB / 16.4% CPU)}	🥇 8,098 _{(69 MiB / 27.4% CPU)}	1,138 _{(87 MiB / 27.8% CPU)}	5,599 _{(204 MiB / 41.9% CPU)}
HTTP/3 · QUIC	HTTP/2 · plain	6,267 _{(97 MiB / 45% CPU)}	Not possible (no H2 upstream)	🥇 8,644 _{(67 MiB / 21.8% CPU)}	2,182 _{(83 MiB / 31.8% CPU)}	6,180 _{(179 MiB / 38.4% CPU)}
HTTP/3 · QUIC	HTTP/2 · TLS	5,881 _{(107 MiB / 46% CPU)}	Not possible (no H2 upstream)	🥇 8,089 _{(69 MiB / 24.6% CPU)}	2,252 _{(84 MiB / 30.9% CPU)}	7,557 _{(171 MiB / 39.6% CPU)}
HTTP/3 · QUIC	HTTP/3 · QUIC	3,988 _{(97 MiB / 39.4% CPU)}	Not possible (no H3 upstream)	Not possible (no H3 upstream)	904 _{(81 MiB / 27.3% CPU)}	🥇 4,700 _{(182 MiB / 40.8% CPU)}

MITM (TWP only)

Same Client×Origin wires with interception on (compare-product 35092827644). Lite = no-op handlers (unchanged-lite finish). Full = append-only header mutation (harness: one probe header each way; product: generic append-only relay via MitmCompressedRelayHelper). nginx/HAProxy/Envoy/YARP cannot MITM. Lite÷Reverse / Full÷Reverse vs bare reverse (same job / comparison-group shard). Completion gate: Lite ≥ 0.50× and Full ≥ 0.50× reverse sustain @ c=64 (median of 3 GHA runs); reverse TWP÷YARP ≥ 0.75× (no terminate-peer gate).

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

v2 drop-only + non-unique append (2026-08-27): MitmStaticRebuildHelper rebuilds static HPACK/QPACK after 1–4 unique header drops; trailing non-unique appends stay on compressed relay. @ 8bfa7852: 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	13,180 _{(84 MiB / 39.4% CPU)}	13,546 _{(85 MiB / 38.2% CPU)}	0.81×	0.83×
HTTP/1 · plain	HTTP/1 · TLS	11,128 _{(95 MiB / 40% CPU)}	10,009 _{(94 MiB / 37.9% CPU)}	0.87×	0.78×
HTTP/1 · plain	HTTP/2 · plain	18,046 _{(91 MiB / 44.7% CPU)}	12,264 _{(93 MiB / 43.8% CPU)}	0.86×	0.59×
HTTP/1 · plain	HTTP/2 · TLS	12,795 _{(106 MiB / 40.2% CPU)}	17,562 _{(106 MiB / 42.6% CPU)}	0.96×	1.32×
HTTP/1 · plain	HTTP/3 · QUIC	4,402 _{(91 MiB / 47.8% CPU)}	4,047 _{(93 MiB / 45.6% CPU)}	0.94×	0.86×
HTTP/1 · TLS	HTTP/1 · plain	11,591 _{(95 MiB / 39.4% CPU)}	7,671 _{(96 MiB / 38.9% CPU)}	1.46×	0.96×
HTTP/1 · TLS	HTTP/1 · TLS	8,412 _{(99 MiB / 37.9% CPU)}	10,725 _{(99 MiB / 38.9% CPU)}	0.94×	1.19×
HTTP/1 · TLS	HTTP/2 · plain	10,748 _{(100 MiB / 41.7% CPU)}	9,978 _{(105 MiB / 39.5% CPU)}	1.09×	1.01×
HTTP/1 · TLS	HTTP/2 · TLS	11,660 _{(161 MiB / 39.5% CPU)}	6,751 _{(156 MiB / 37.8% CPU)}	1.48×	0.86×
HTTP/1 · TLS	HTTP/3 · QUIC	2,975 _{(98 MiB / 54.7% CPU)}	3,009 _{(103 MiB / 49.8% CPU)}	1.1×	1.11×
HTTP/2 · plain	HTTP/1 · plain	16,657 _{(90 MiB / 43.4% CPU)}	16,350 _{(90 MiB / 44.4% CPU)}	1.02×	1×
HTTP/2 · plain	HTTP/1 · TLS	14,896 _{(98 MiB / 46.9% CPU)}	11,821 _{(97 MiB / 47% CPU)}	1.2×	0.95×
HTTP/2 · plain	HTTP/2 · plain	28,786 _{(79 MiB / 34.8% CPU)}	29,833 _{(76 MiB / 36.4% CPU)}	0.78×	0.81×
HTTP/2 · plain	HTTP/2 · TLS	26,546 _{(80 MiB / 38.2% CPU)}	34,124 _{(83 MiB / 36.4% CPU)}	0.87×	1.11×
HTTP/2 · plain	HTTP/3 · QUIC	4,078 _{(95 MiB / 52.4% CPU)}	4,254 _{(94 MiB / 53.3% CPU)}	0.8×	0.84×
HTTP/2 · TLS	HTTP/1 · plain	15,130 _{(95 MiB / 48.3% CPU)}	15,199 _{(92 MiB / 45.1% CPU)}	0.95×	0.96×
HTTP/2 · TLS	HTTP/1 · TLS	15,237 _{(100 MiB / 41.4% CPU)}	13,176 _{(97 MiB / 47% CPU)}	1.14×	0.99×
HTTP/2 · TLS	HTTP/2 · plain	26,045 _{(87 MiB / 38% CPU)}	31,311 _{(86 MiB / 36.2% CPU)}	0.95×	1.14×
HTTP/2 · TLS	HTTP/2 · TLS	29,113 _{(89 MiB / 36.4% CPU)}	28,038 _{(90 MiB / 38.6% CPU)}	0.95×	0.92×
HTTP/2 · TLS	HTTP/3 · QUIC	5,228 _{(107 MiB / 46.1% CPU)}	3,875 _{(107 MiB / 45.1% CPU)}	0.92×	0.68×
HTTP/3 · QUIC	HTTP/1 · plain	4,428 _{(101 MiB / 45.8% CPU)}	5,675 _{(102 MiB / 47.5% CPU)}	0.96×	1.23×
HTTP/3 · QUIC	HTTP/1 · TLS	4,527 _{(116 MiB / 46.2% CPU)}	4,613 _{(116 MiB / 44.1% CPU)}	1.06×	1.08×
HTTP/3 · QUIC	HTTP/2 · plain	5,291 _{(99 MiB / 49.4% CPU)}	4,946 _{(99 MiB / 48.2% CPU)}	0.84×	0.79×
HTTP/3 · QUIC	HTTP/2 · TLS	4,271 _{(105 MiB / 46.1% CPU)}	4,538 _{(107 MiB / 48.8% CPU)}	0.73×	0.77×
HTTP/3 · QUIC	HTTP/3 · QUIC	3,764 _{(95 MiB / 57.2% CPU)}	4,233 _{(93 MiB / 41.9% CPU)}	0.94×	1.06×

Editions (CLI / Plus / Intercept)

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

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

Arm	Win	Linux	Win÷	Lin÷	Gate
`twp-cli-reverse-http1` vs library	32,214 _{(peak 32,219 · 118 MiB / 43.4% CPU)}	47,734 _{(peak 47,844 · 151 MiB / 49.0% CPU)}	1.02×	1.04×	≥ 0.80×
`twp-cli-reverse-http1-tls` vs library TLS	26,495 _{(peak 26,651 · 138 MiB / 48.3% CPU)}	36,932 _{(peak 37,200 · 174 MiB / 48.5% CPU)}	1.01×	1.00×	≥ 0.80×
`twp-cli-reverse-http1-route` vs CLI	32,394 _{(peak 32,428 · 120 MiB / 47.7% CPU)}	47,369 _{(peak 47,398 · 149 MiB / 49.2% CPU)}	1.01×	0.99×	≥ 0.90×
`twp-cli-plus-base-http1` vs CLI	32,475 _{(peak 32,695 · 123 MiB / 46.0% CPU)}	47,322 _{(peak 47,744 · 154 MiB / 49.8% CPU)}	1.01×	0.99×	≥ 0.90×
`twp-cli-plus-cache-http1` (cold) vs CLI	34,029 _{(peak 34,460 · 118 MiB / 64.5% CPU)}	49,409 _{(peak 50,104 · 148 MiB / 62.3% CPU)}	1.06×	1.04×	≥ 0.70×
`twp-cli-intercept-http1` vs CLI	25,219 _{(peak 25,377 · 124 MiB / 54.0% CPU)}	35,969 _{(peak 35,977 · 155 MiB / 51.2% CPU)}	0.78×	0.75×	≥ 0.70×
`twp-cli-plus-waf-http1` vs CLI	31,653 _{(peak 31,837 · 124 MiB / 46.8% CPU)}	46,367 _{(peak 46,685 · 151 MiB / 49.6% CPU)}	0.98×	0.97×	≥ 0.80×
`twp-cli-plus-cidr-http1` vs CLI	31,730 _{(peak 31,757 · 123 MiB / 51.1% CPU)}	46,828 _{(peak 47,093 · 150 MiB / 50.2% CPU)}	0.98×	0.98×	≥ 0.80×
`twp-cli-plus-jwt-http1` vs CLI	30,326 _{(peak 30,386 · 138 MiB / 46.6% CPU)}	43,906 _{(peak 44,433 · 179 MiB / 49.5% CPU)}	0.94×	0.92×	≥ 0.70×
`twp-cli-plus-ratelimit-http1` vs CLI	31,713 _{(peak 31,944 · 123 MiB / 48.8% CPU)}	46,435 _{(peak 46,860 · 151 MiB / 49.4% CPU)}	0.98×	0.97×	≥ 0.80×
`twp-cli-plus-resilience-http1` vs CLI	32,373 _{(peak 32,416 · 128 MiB / 50.5% CPU)}	47,302 _{(peak 47,654 · 155 MiB / 49.5% CPU)}	1.00×	0.99×	≥ 0.85×
`twp-cli-plus-discovery-file-http1` vs CLI	32,209 _{(peak 32,332 · 121 MiB / 46.5% CPU)}	47,485 _{(peak 47,770 · 151 MiB / 49.1% CPU)}	1.00×	0.99×	≥ 0.80×
`twp-cli-plus-metrics-scrape-http1` vs CLI	32,192 _{(peak 32,427 · 126 MiB / 44.8% CPU)}	47,348 _{(peak 48,665 · 159 MiB / 49.5% CPU)}	1.00×	0.99×	≥ 0.80×
`twp-cli-plus-cache-hit-http1` vs cache cold	34,015 _{(peak 34,022 · 117 MiB / 66.0% CPU)}	49,153 _{(peak 49,358 · 149 MiB / 63.1% CPU)}	1.00×	0.99×	≥ 0.90×
`twp-cli-static-http1` vs CLI	53,078 _{(peak 53,980 · 109 MiB / 52.1% CPU)}	78,333 _{(peak 82,955 · 150 MiB / 48.4% CPU)}	1.65×	1.64×	≥ 0.85×
`twp-cli-logging-http1` vs CLI	32,440 _{(peak 32,581 · 119 MiB / 47.4% CPU)}	47,624 _{(peak 48,604 · 150 MiB / 49.3% CPU)}	1.01×	1.00×	≥ 0.90×
`twp-cli-lb-leasttime-http1` vs route	29,649 _{(peak 30,386 · 134 MiB / 54.7% CPU)}	44,790 _{(peak 44,825 · 180 MiB / 50.6% CPU)}	0.92×	0.95×	≥ 0.85×
`twp-cli-dialect-twp-http1` vs CLI	32,461 _{(peak 32,780 · 114 MiB / 49.7% CPU)}	47,898 _{(peak 48,154 · 145 MiB / 49.3% CPU)}	1.01×	1.00×	≥ 0.90×

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

Cross-version (7.0 vs 6.0)

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

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

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

Heavier reverse workloads

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

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

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

Median of 3 repeats on windows-latest @ 8bfa7852. Source: Actions 35092832912 (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	🥇 8,782 _{(114 MiB / 46.0% CPU)}	644 _{(142 MiB / 24.7% CPU)}	7,764 _{(135 MiB / 47.3% CPU)}
64 KiB	HTTP/2 · TLS	HTTP/1 · plain	🥇 8,184 _{(174 MiB / 46.0% CPU)}	564 _{(142 MiB / 24.8% CPU)}	7,013 _{(135 MiB / 50.5% CPU)}
64 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 3,876 _{(132 MiB / 38.6% CPU)}	Not possible (no QUIC)	3,663 _{(188 MiB / 46.8% CPU)}
256 KiB	HTTP/1 · TLS	HTTP/1 · plain	🥇 2,790 _{(129 MiB / 47.6% CPU)}	0 _{(peak 162 · 143 MiB / 24.8% CPU)}	2,586 _{(135 MiB / 48.0% CPU)}
256 KiB	HTTP/2 · TLS	HTTP/1 · plain	🥇 2,537 _{(153 MiB / 40.3% CPU)}	0 _{(peak 142 · 142 MiB / 24.7% CPU)}	1,909 _{(134 MiB / 45.1% CPU)}
256 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 1,121 _{(107 MiB / 39.8% CPU)}	Not possible (no QUIC)	1,048 _{(170 MiB / 46.1% CPU)}
64 KiB	HTTP/2 · plain	HTTP/1 · plain	🥇 11,532 _{(170 MiB / 42.4% CPU)}	2,087 _{(127 MiB / 24.8% CPU)}	9,578 _{(118 MiB / 45.0% CPU)}
64 KiB	HTTP/2 · TLS	HTTP/2 · plain	3,848 _{(77 MiB / 35.2% CPU)}	Not possible	🥇 6,673 _{(140 MiB / 47.9% CPU)}
64 KiB	HTTP/2 · TLS	HTTP/2 · TLS	3,495 _{(78 MiB / 37.3% CPU)}	Not possible	🥇 6,000 _{(140 MiB / 47.2% CPU)}
64 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	3,396 _{(125 MiB / 43.6% CPU)}	Not possible	🥇 3,552 _{(184 MiB / 46.8% CPU)}
64 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	🥇 3,934 _{(138 MiB / 41.5% CPU)}	Not possible (no QUIC)	3,307 _{(199 MiB / 49.0% CPU)}
256 KiB	HTTP/2 · plain	HTTP/1 · plain	🥇 3,784 _{(144 MiB / 30.6% CPU)}	655 _{(127 MiB / 24.3% CPU)}	2,823 _{(123 MiB / 38.1% CPU)}
256 KiB	HTTP/2 · TLS	HTTP/2 · plain	1,382 _{(86 MiB / 33.1% CPU)}	Not possible	🥇 1,808 _{(163 MiB / 47.4% CPU)}
256 KiB	HTTP/2 · TLS	HTTP/2 · TLS	1,277 _{(87 MiB / 34.7% CPU)}	Not possible	🥇 1,410 _{(146 MiB / 44.2% CPU)}
256 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	869 _{(151 MiB / 42.2% CPU)}	Not possible	🥇 979 _{(191 MiB / 44.0% CPU)}
256 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	🥇 1,090 _{(148 MiB / 40.6% CPU)}	Not possible (no QUIC)	990 _{(198 MiB / 44.3% CPU)}

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

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

Median of 3 repeats @ 8bfa7852. Source: Actions 35092832912 (compare-bodies). Warmup 2s / measure 8s.

Body	Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
64 KiB	HTTP/1 · TLS	HTTP/1 · plain	8,545 _{(175 MiB / 44.2% CPU)}	5,990 _{(100 MiB / 49.8% CPU)}	🥇 9,584 _{(83 MiB / 39.5% CPU)}	8,284 _{(132 MiB / 43.3% CPU)}	7,035 _{(164 MiB / 47.9% CPU)}
64 KiB	HTTP/2 · TLS	HTTP/1 · plain	🥇 5,771 _{(221 MiB / 40.2% CPU)}	1,831 _{(101 MiB / 16.9% CPU)}	4,786 _{(84 MiB / 24.4% CPU)}	4,353 _{(140 MiB / 23.6% CPU)}	4,800 _{(162 MiB / 48.5% CPU)}
64 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 5,873 _{(183 MiB / 43.5% CPU)}	0 _{(peak 1,851 · 129 MiB / 23.6% CPU)}	5,030 _{(88 MiB / 32.8% CPU)}	123 _{(139 MiB / 1.1% CPU)}	4,568 _{(231 MiB / 51.3% CPU)}
256 KiB	HTTP/1 · TLS	HTTP/1 · plain	2,794 _{(129 MiB / 35.7% CPU)}	1,514 _{(100 MiB / 47.3% CPU)}	🥇 3,024 _{(83 MiB / 32.8% CPU)}	2,727 _{(145 MiB / 32.2% CPU)}	2,211 _{(170 MiB / 44.0% CPU)}
256 KiB	HTTP/2 · TLS	HTTP/1 · plain	1,563 _{(195 MiB / 37.2% CPU)}	595 _{(100 MiB / 21.3% CPU)}	1,588 _{(83 MiB / 21.1% CPU)}	🥇 1,900 _{(158 MiB / 20.8% CPU)}	1,355 _{(161 MiB / 45.0% CPU)}
256 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 1,519 _{(159 MiB / 42.6% CPU)}	0 _{(118 MiB / 20.6% CPU)}	1,344 _{(89 MiB / 30.3% CPU)}	25 _{(157 MiB / 0.7% CPU)}	1,304 _{(218 MiB / 47.4% CPU)}
64 KiB	HTTP/2 · plain	HTTP/1 · plain	🥇 9,681 _{(237 MiB / 42.5% CPU)}	2,321 _{(80 MiB / 12.9% CPU)}	7,167 _{(70 MiB / 24.3% CPU)}	6,374 _{(132 MiB / 23.9% CPU)}	8,267 _{(153 MiB / 45.9% CPU)}
64 KiB	HTTP/2 · TLS	HTTP/2 · plain	3,709 _{(103 MiB / 38.7% CPU)}	Not possible	2,594 _{(86 MiB / 24.7% CPU)}	🥇 5,209 _{(144 MiB / 22.1% CPU)}	4,981 _{(184 MiB / 46.5% CPU)}
64 KiB	HTTP/2 · TLS	HTTP/2 · TLS	2,716 _{(102 MiB / 38.4% CPU)}	Not possible	1,584 _{(83 MiB / 24.5% CPU)}	3,262 _{(146 MiB / 23.0% CPU)}	🥇 3,359 _{(171 MiB / 45.6% CPU)}
64 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	3,060 _{(164 MiB / 53.0% CPU)}	Not possible	2,380 _{(90 MiB / 35.2% CPU)}	1 _{(144 MiB / 0.1% CPU)}	🥇 3,219 _{(240 MiB / 49.1% CPU)}
64 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	🥇 3,926 _{(203 MiB / 44.0% CPU)}	0 _{(peak 1,472 · 143 MiB / 23.0% CPU)}	3,080 _{(94 MiB / 31.8% CPU)}	0 _{(140 MiB / 0.1% CPU)}	3,182 _{(240 MiB / 49.5% CPU)}
256 KiB	HTTP/2 · plain	HTTP/1 · plain	2,768 _{(211 MiB / 36.3% CPU)}	877 _{(79 MiB / 20.1% CPU)}	2,746 _{(70 MiB / 22.6% CPU)}	🥇 3,006 _{(152 MiB / 22.6% CPU)}	2,859 _{(157 MiB / 38.6% CPU)}
256 KiB	HTTP/2 · TLS	HTTP/2 · plain	1,203 _{(113 MiB / 33.7% CPU)}	Not possible	772 _{(90 MiB / 24.3% CPU)}	🥇 1,769 _{(169 MiB / 22.0% CPU)}	1,350 _{(187 MiB / 41.8% CPU)}
256 KiB	HTTP/2 · TLS	HTTP/2 · TLS	804 _{(109 MiB / 33.0% CPU)}	Not possible	433 _{(82 MiB / 24.6% CPU)}	🥇 983 _{(162 MiB / 22.5% CPU)}	954 _{(179 MiB / 41.9% CPU)}
256 KiB	HTTP/3 · QUIC	HTTP/2 · TLS	847 _{(193 MiB / 51.2% CPU)}	Not possible	706 _{(93 MiB / 36.1% CPU)}	0 _{(165 MiB / 0.1% CPU)}	🥇 948 _{(230 MiB / 46.3% CPU)}
256 KiB	HTTP/3 · QUIC	HTTP/1 · TLS	🥇 1,090 _{(208 MiB / 45.0% CPU)}	0 _{(peak 216 · 144 MiB / 21.9% CPU)}	872 _{(96 MiB / 32.0% CPU)}	0 _{(148 MiB / 0.2% CPU)}	989 _{(245 MiB / 47.6% CPU)}

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

Windows — POST 64 KiB request + 64 KiB response

Median of 3 repeats on windows-latest @ 8bfa7852. Source: Actions 35092838467 (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	🥇 6,265 _{(98 MiB / 43.7% CPU)}	361 _{(142 MiB / 24.7% CPU)}	4,403 _{(137 MiB / 54.9% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	🥇 4,552 _{(190 MiB / 48.0% CPU)}	364 _{(144 MiB / 24.9% CPU)}	2,959 _{(134 MiB / 38.4% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	🥇 2,026 _{(179 MiB / 39.5% CPU)}	Not possible (no QUIC)	1,928 _{(218 MiB / 49.6% CPU)}
HTTP/2 · plain	HTTP/1 · plain	🥇 6,752 _{(189 MiB / 45.0% CPU)}	1,799 _{(130 MiB / 24.7% CPU)}	6,174 _{(123 MiB / 52.5% CPU)}
HTTP/2 · TLS	HTTP/2 · plain	8 _{(88 MiB / 0.2% CPU)}	Not possible	🥇 3,734 _{(145 MiB / 47.7% CPU)}
HTTP/2 · TLS	HTTP/2 · TLS	8 _{(89 MiB / 0.1% CPU)}	Not possible	🥇 2,795 _{(145 MiB / 46.3% CPU)}
HTTP/3 · QUIC	HTTP/2 · TLS	1,801 _{(180 MiB / 45.2% CPU)}	Not possible	🥇 1,870 _{(209 MiB / 49.3% CPU)}
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 1,930 _{(180 MiB / 41.5% CPU)}	Not possible (no QUIC)	1,830 _{(221 MiB / 48.5% CPU)}

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

Linux — POST 64 KiB request + 64 KiB response

Median of 3 repeats @ 8bfa7852. Source: Actions 35092838467 (compare-post).

Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
HTTP/1 · TLS	HTTP/1 · plain	4,706 _{(131 MiB / 45.2% CPU)}	3,494 _{(100 MiB / 48.9% CPU)}	4,766 _{(85 MiB / 41.1% CPU)}	🥇 4,816 _{(133 MiB / 42.2% CPU)}	3,114 _{(174 MiB / 55.3% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	🥇 3,051 _{(210 MiB / 47.8% CPU)}	1,296 _{(113 MiB / 21.6% CPU)}	1,773 _{(82 MiB / 24.1% CPU)}	0 _{(peak 2,797 · 142 MiB / 22.5% CPU)}	2,483 _{(170 MiB / 48.6% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	🥇 2,881 _{(226 MiB / 44.1% CPU)}	457 _{(109 MiB / 25.0% CPU)}	2,078 _{(91 MiB / 35.9% CPU)}	0 _{(125 MiB / 0.1% CPU)}	2,552 _{(259 MiB / 49.6% CPU)}
HTTP/2 · plain	HTTP/1 · plain	🥇 5,393 _{(226 MiB / 44.3% CPU)}	2,294 _{(97 MiB / 22.5% CPU)}	2,629 _{(69 MiB / 23.7% CPU)}	0 _{(peak 4,450 · 133 MiB / 23.6% CPU)}	4,330 _{(160 MiB / 48.2% CPU)}
HTTP/2 · TLS	HTTP/2 · plain	2 _{(117 MiB / 0.1% CPU)}	Not possible	1,363 _{(86 MiB / 24.4% CPU)}	🥇 2,826 _{(141 MiB / 23.5% CPU)}	2,367 _{(181 MiB / 47.3% CPU)}
HTTP/2 · TLS	HTTP/2 · TLS	8 _{(121 MiB / 0.3% CPU)}	Not possible	1,048 _{(87 MiB / 24.5% CPU)}	🥇 2,023 _{(147 MiB / 23.1% CPU)}	1,880 _{(175 MiB / 46.6% CPU)}
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 1,834 _{(240 MiB / 49.0% CPU)}	Not possible	1,118 _{(95 MiB / 31.6% CPU)}	0 _{(126 MiB / 0.1% CPU)}	1,761 _{(255 MiB / 47.1% CPU)}
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 2,088 _{(238 MiB / 44.9% CPU)}	367 _{(120 MiB / 24.9% CPU)}	1,492 _{(95 MiB / 32.4% CPU)}	0 _{(126 MiB / 0.1% CPU)}	1,944 _{(257 MiB / 47.9% CPU)}

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

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

Userspace 5 ms one-way delay + 1% TCP connection stall (H1/H2) or UDP datagram drop (H3); 64 KiB GET. Median of 3 repeats on windows-latest @ 8bfa7852 — 35092841035 (compare-lossy).

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

Client	Origin	TWP	nginx	YARP
HTTP/1 · TLS	HTTP/1 · plain	663 _{(112 MiB / 4.2% CPU)}	653 _{(142 MiB / 16.7% CPU)}	🥇 679 _{(123 MiB / 3.9% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	0 _{(peak 89 · 121 MiB / 1.4% CPU)}	0 _{(peak 17 · 142 MiB / 0.5% CPU)}	0 _{(peak 16 · 100 MiB / 0.7% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	0 _{(67 MiB / 0.0% CPU)}	Not possible (no QUIC)	0 _{(79 MiB / 0.0% CPU)}
HTTP/2 · plain	HTTP/1 · plain	0 _{(peak 88 · 131 MiB / 0.9% CPU)}	0 _{(peak 17 · 128 MiB / 0.1% CPU)}	0 _{(peak 16 · 91 MiB / 0.7% CPU)}
HTTP/2 · TLS	HTTP/2 · plain	0 _{(peak 8 · 67 MiB / 0.5% CPU)}	Not possible	0 _{(peak 16 · 101 MiB / 0.9% CPU)}
HTTP/2 · TLS	HTTP/2 · TLS	0 _{(peak 9 · 72 MiB / 0.3% CPU)}	Not possible	0 _{(peak 15 · 100 MiB / 0.6% CPU)}
HTTP/3 · QUIC	HTTP/2 · TLS	0 _{(71 MiB / 0.0% CPU)}	Not possible	0 _{(80 MiB / 0.0% CPU)}
HTTP/3 · QUIC	HTTP/1 · TLS	0 _{(69 MiB / 0.0% CPU)}	Not possible (no QUIC)	0 _{(79 MiB / 0.0% CPU)}

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

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

Median of 3 repeats @ 8bfa7852. Source: 35092841035 (compare-lossy; lossy H3 uses quic-http3).

Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
HTTP/1 · TLS	HTTP/1 · plain	1,214 _{(148 MiB / 10.2% CPU)}	🥇 1,219 _{(102 MiB / 8.2% CPU)}	1,218 _{(83 MiB / 4.8% CPU)}	1,209 _{(128 MiB / 6.6% CPU)}	1,211 _{(146 MiB / 13.1% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	🥇 312 _{(186 MiB / 5.2% CPU)}	40 _{(101 MiB / 0.2% CPU)}	40 _{(84 MiB / 0.2% CPU)}	40 _{(137 MiB / 0.3% CPU)}	40 _{(129 MiB / 1.1% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	342 _{(158 MiB / 10.4% CPU)}	100 _{(112 MiB / 1.8% CPU)}	60 _{(86 MiB / 2.7% CPU)}	🥇 1,578 _{(140 MiB / 20.3% CPU)}	321 _{(178 MiB / 15.2% CPU)}
HTTP/2 · plain	HTTP/1 · plain	🥇 337 _{(177 MiB / 5.2% CPU)}	40 _{(78 MiB / 0.1% CPU)}	41 _{(67 MiB / 0.1% CPU)}	40 _{(127 MiB / 0.3% CPU)}	41 _{(123 MiB / 1.0% CPU)}
HTTP/2 · TLS	HTTP/2 · plain	0 _{(peak 16 · 97 MiB / 0.5% CPU)}	Not possible	🥇 42 _{(90 MiB / 0.3% CPU)}	40 _{(136 MiB / 0.2% CPU)}	41 _{(133 MiB / 1.0% CPU)}
HTTP/2 · TLS	HTTP/2 · TLS	0 _{(peak 13 · 98 MiB / 0.5% CPU)}	Not possible	🥇 41 _{(84 MiB / 0.4% CPU)}	40 _{(138 MiB / 0.4% CPU)}	41 _{(131 MiB / 1.2% CPU)}
HTTP/3 · QUIC	HTTP/2 · TLS	325 _{(143 MiB / 19.3% CPU)}	Not possible	66 _{(91 MiB / 3.2% CPU)}	🥇 1,459 _{(145 MiB / 23.4% CPU)}	349 _{(190 MiB / 20.2% CPU)}
HTTP/3 · QUIC	HTTP/1 · TLS	325 _{(165 MiB / 10.9% CPU)}	99 _{(118 MiB / 1.9% CPU)}	63 _{(90 MiB / 3.1% CPU)}	🥇 1,456 _{(141 MiB / 22.1% CPU)}	350 _{(185 MiB / 18.6% CPU)}

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

Architecture-sensitive

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

Median of 3 repeats on matched 4 vCPU / 16 GiB runners @ 8bfa7852 (35092846209). 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	🥇 248 _{(100 MiB / 4.9% CPU)}	192 _{(143 MiB / 24.7% CPU)}	240 _{(111 MiB / 4.6% CPU)}
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/1 · plain	248 _{(121 MiB / 4.6% CPU)}	180 _{(142 MiB / 24.7% CPU)}	🥇 253 _{(112 MiB / 6.7% CPU)}
Slow consumer (256 KiB GET, throttled client read)	HTTP/3 · QUIC	HTTP/1 · plain	273 _{(102 MiB / 15.2% CPU)}	Not possible (no QUIC)	🥇 274 _{(162 MiB / 15.7% CPU)}
Early response (origin writes after first request chunk)	HTTP/1 · TLS	HTTP/1 · plain	🥇 11,064 _{(101 MiB / 41.4% CPU)}	679 _{(143 MiB / 24.9% CPU)}	7,558 _{(138 MiB / 53.7% CPU)}
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/1 · plain	2,870 _{(181 MiB / 40.4% CPU)}	0 _{(peak 343 · 144 MiB / 24.9% CPU)}	🥇 3,336 _{(135 MiB / 51.9% CPU)}
Early response (origin writes after first request chunk)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 2,141 _{(149 MiB / 41.8% CPU)}	Not possible (no QUIC)	1,888 _{(199 MiB / 52.7% CPU)}
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · plain	HTTP/1 · plain	248 _{(106 MiB / 3.9% CPU)}	248 _{(127 MiB / 8.7% CPU)}	🥇 256 _{(104 MiB / 4.3% CPU)}
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/2 · TLS	0 _{(peak 8 · 72 MiB / 0.3% CPU)}	Not possible	🥇 256 _{(141 MiB / 4.9% CPU)}
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/2 · TLS	0 _{(97 MiB / 0.1% CPU)}	Not possible	0 _{(123 MiB / 0.2% CPU)}
Duplex (both directions live)	HTTP/2 · TLS	HTTP/2 · TLS	0 _{(95 MiB / 0.1% CPU)}	Not possible	0 _{(114 MiB / 0.2% CPU)}
Duplex (WebSocket / H1 Upgrade)	HTTP/1 · TLS	HTTP/1 · plain	🥇 66,034 _{(97 MiB / 41.3% CPU)}	36,248 _{(143 MiB / 24.6% CPU)}	61,591 _{(89 MiB / 44.1% CPU)}

Linux

Scenario	Client	Origin	TWP	nginx	HAProxy	Envoy	YARP
Slow consumer (256 KiB GET, throttled client read)	HTTP/1 · TLS	HTTP/1 · plain	468 _{(125 MiB / 9.2% CPU)}	416 _{(101 MiB / 9.4% CPU)}	🥇 472 _{(84 MiB / 5.8% CPU)}	469 _{(139 MiB / 6.2% CPU)}	419 _{(145 MiB / 14.0% CPU)}
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/1 · plain	473 _{(157 MiB / 12.2% CPU)}	427 _{(102 MiB / 9.4% CPU)}	474 _{(84 MiB / 5.4% CPU)}	🥇 480 _{(157 MiB / 5.0% CPU)}	474 _{(148 MiB / 16.2% CPU)}
Slow consumer (256 KiB GET, throttled client read)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 480 _{(135 MiB / 31.0% CPU)}	0 _{(peak 347 · 124 MiB / 18.5% CPU)}	479 _{(90 MiB / 7.8% CPU)}	1 _{(156 MiB / 0.2% CPU)}	476 _{(196 MiB / 37.0% CPU)}
Early response (origin writes after first request chunk)	HTTP/1 · TLS	HTTP/1 · plain	6,746 _{(148 MiB / 44.2% CPU)}	5,410 _{(101 MiB / 48.0% CPU)}	🥇 7,255 _{(84 MiB / 39.9% CPU)}	0 _{(peak 673 · 130 MiB / 6.6% CPU)}	4,645 _{(176 MiB / 55.1% CPU)}
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/1 · plain	2,465 _{(210 MiB / 48.0% CPU)}	0 _{(peak 858 · 112 MiB / 15.2% CPU)}	🥇 2,509 _{(85 MiB / 24.7% CPU)}	0 _{(peak 2,574 · 145 MiB / 23.1% CPU)}	2,226 _{(171 MiB / 48.6% CPU)}
Early response (origin writes after first request chunk)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 4,647 _{(224 MiB / 46.4% CPU)}	0 _{(peak 706 · 122 MiB / 24.5% CPU)}	2,866 _{(93 MiB / 32.9% CPU)}	0 _{(126 MiB / 0.1% CPU)}	3,107 _{(258 MiB / 47.4% CPU)}
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · plain	HTTP/1 · plain	475 _{(139 MiB / 14.4% CPU)}	463 _{(79 MiB / 11.5% CPU)}	🥇 476 _{(68 MiB / 6.2% CPU)}	472 _{(150 MiB / 4.5% CPU)}	474 _{(149 MiB / 14.7% CPU)}
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/2 · TLS	9 _{(96 MiB / 1.8% CPU)}	Not possible	447 _{(90 MiB / 12.9% CPU)}	470 _{(157 MiB / 9.3% CPU)}	🥇 470 _{(172 MiB / 21.4% CPU)}
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/2 · TLS	0 _{(117 MiB / 0.1% CPU)}	Not possible	🥇 8 _{(95 MiB / 0.2% CPU)}	0 _{(peak 2,844 · 152 MiB / 19.7% CPU)}	0 _{(141 MiB / 0.1% CPU)}
Duplex (both directions live)	HTTP/2 · TLS	HTTP/2 · TLS	0 _{(118 MiB / 0.1% CPU)}	Not possible	0 _{(93 MiB / 0.1% CPU)}	0 _{(peak 2,526 · 156 MiB / 17.2% CPU)}	0 _{(146 MiB / 0.2% CPU)}
Duplex (WebSocket / H1 Upgrade)	HTTP/1 · TLS	HTTP/1 · plain	45,832 _{(128 MiB / 44.4% CPU)}	🥇 49,425 _{(102 MiB / 36.9% CPU)}	48,657 _{(82 MiB / 39.8% CPU)}	46,062 _{(127 MiB / 41.2% CPU)}	41,388 _{(126 MiB / 45.3% CPU)}

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

TLS termination cost (H1 TLS → cleartext origin)

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

Windows

Median of 3 repeats on windows-latest @ 8bfa7852. Source: Actions 35092843724. 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	🥇 21,760 _{(85 MiB / 49.0% CPU)}	9,611 _{(142 MiB / 24.8% CPU)}	19,339 _{(101 MiB / 48.2% CPU)}
New-connection · tiny GET	🥇 747 _{(89 MiB / 9.4% CPU)}	0 _{(peak 247 · 140 MiB / 24.2% CPU)}	732 _{(113 MiB / 9.5% CPU)}
Keep-alive · 256 KiB GET	🥇 2,948 _{(133 MiB / 45.9% CPU)}	0 _{(peak 158 · 143 MiB / 24.5% CPU)}	2,718 _{(128 MiB / 45.9% CPU)}

Linux

Median of 3 repeats @ 8bfa7852. Source: Actions 35092843724.

Workload	TWP	nginx	HAProxy	Envoy	YARP
Keep-alive · tiny GET	23,990 _{(112 MiB / 50.2% CPU)}	28,472 _{(100 MiB / 40.9% CPU)}	🥇 28,484 _{(83 MiB / 43.2% CPU)}	17,372 _{(127 MiB / 58.6% CPU)}	20,581 _{(136 MiB / 50.5% CPU)}
New-connection · tiny GET	976 _{(116 MiB / 47.4% CPU)}	1,017 _{(100 MiB / 44.5% CPU)}	956 _{(82 MiB / 44.1% CPU)}	🥇 1,072 _{(129 MiB / 41.4% CPU)}	967 _{(152 MiB / 46.4% CPU)}
Keep-alive · 256 KiB GET	2,708 _{(128 MiB / 36.8% CPU)}	1,740 _{(100 MiB / 53.7% CPU)}	🥇 2,966 _{(84 MiB / 33.2% CPU)}	2,734 _{(146 MiB / 34.2% CPU)}	2,145 _{(172 MiB / 45.3% CPU)}

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

Unary gRPC (H2 TLS)

Unary Echo RPC/s @ c=64 over H2 TLS→H2 TLS for Titanium, YARP, nginx (grpc_pass), HAProxy, and Envoy (OS-possible peers only). Not folded into the architecture-sensitive tables. Median of 3 repeats @ 8bfa7852 — 35092856793.

OS	Titanium	YARP	nginx	HAProxy	Envoy
Windows	111,791 _{(116 MiB / 22.8% CPU)}	63,757 _{(120 MiB / 41.9% CPU)}	0 _{(143 MiB / 24.9% CPU)}	Not possible	Not possible
Linux	78,834 _{(121 MiB / 32.5% CPU)}	45,952 _{(149 MiB / 42.2% CPU)}	0 _{(131 MiB / 25.1% CPU)}	18,339 _{(83 MiB / 24.2% CPU)}	30,733 _{(127 MiB / 21.4% CPU)}
macOS	18,711 _{(97 MiB / 24.0% CPU)}	13,574 _{(137 MiB / 34.5% CPU)}	0 _{(61 MiB / 3.1% CPU)}	0 _{(63 MiB / 18.4% CPU)}	7,199 _{(80 MiB / 20.7% CPU)}

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 @ 8bfa7852 — 35092859116.

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

OS	Titanium	YARP	HAProxy	Envoy
Windows	60,626 _{(93 MiB / 24.4% CPU)}	34,219 _{(117 MiB / 47.2% CPU)}	Not possible	Not possible
Linux	43,638 _{(126 MiB / 28.5% CPU)}	24,134 _{(169 MiB / 43.3% CPU)}	7,783 _{(82 MiB / 24.4% CPU)}	12,047 _{(128 MiB / 22.2% CPU)}
macOS	24,524 _{(95 MiB / 24.8% CPU)}	16,722 _{(130 MiB / 35.8% CPU)}	0 _{(65 MiB / 17.4% CPU)}	0 _{(79 MiB / 20.6% CPU)}

WebSocket (H1 TLS → H1 TLS)

WebSocket echo round-trips/sec over dual-TLS H1 (proxy_ssl style). Mode: compare-ws-h1tls (*-duplex-ws-h1tls).

OS	Titanium	YARP	nginx	HAProxy	Envoy
Windows	66,023 _{(102 MiB / 42.5% CPU)}	63,442 _{(90 MiB / 46.5% CPU)}	31,084 _{(143 MiB / 24.7% CPU)}	Not possible	Not possible
Linux	38,814 _{(137 MiB / 44.3% CPU)}	34,209 _{(136 MiB / 45.8% CPU)}	41,105 _{(108 MiB / 38.2% CPU)}	42,059 _{(85 MiB / 39.8% CPU)}	40,348 _{(127 MiB / 40.1% CPU)}
macOS	10,520 _{(144 MiB / 33.3% CPU)}	8,988 _{(145 MiB / 33.6% CPU)}	10,184 _{(81 MiB / 26.4% CPU)}	10,985 _{(67 MiB / 32.0% CPU)}	15,732 _{(81 MiB / 30.8% CPU)}

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).

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

OS	Titanium	YARP	HAProxy	Envoy
Windows	23,244 _{(172 MiB / 41.7% CPU)}	0 _{(165 MiB / 19.8% CPU)}	Not possible	Not possible
Linux	1,530 _{(143 MiB / 16.9% CPU)}	0 _{(172 MiB / 25.2% CPU)}	1,535 _{(82 MiB / 5.4% CPU)}	0 _{(123 MiB / 0.3% CPU)}
macOS	4,262 _{(219 MiB / 29.2% CPU)}	0 _{(86 MiB / 12.1% CPU)}	4,087 _{(66 MiB / 35.2% CPU)}	0 _{(73 MiB / 0.6% 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`	~60 min (CLI / Plus / Intercept stress arms)
Cross-version (Gate 2)	`compare-cross-version`	~1–2h vs committed 6.0 baselines
Pre-wiki smoke	`compare-product-smoke` (Linux 2 shards, `repeats=1`)	~30–60 min; required before full product
Release / wiki	`compare-product` (3 comparison-group shards × Win/Linux/mac)	~2–2½h wall (Free account queues beyond 20 jobs)
Unary gRPC	`compare-grpc`	Win/Linux/mac; H2↔H2 + H2→h2c (`arm_shard` 1/2, 2/2)
WebSocket dual-TLS	`compare-ws-h1tls`	Win/Linux/mac; H1 TLS→H1 TLS echo
WebSocket RFC 8441	`compare-ws-h2`	Win/Linux/mac; H2 TLS→H1 plain (nginx N/A)
Heavier tables	`compare-bodies` (2 shards) / `post` / `lossy` / `arch` (3 shards) / `tls-cost`	dispatch independently

See PERF-GATES.md.

Arm shards (comparison groups)

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

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

Gate thresholds

Reverse product signal: TWP÷YARP ≥ 0.75 (nginx / HAProxy / Envoy are wiki and chart peers only — no CI gate). Edition CLI/Plus ratios floor at 0.50.
MITM overhead: Lite÷Reverse ≥ 0.50 and Full÷Reverse ≥ 0.50 (median of 3 GHA runs @ c=64). Absolute RPS moves with runner heat; ratios are the claim.
Editions: see PERF-GATES.md and validate-edition-gates.ps1.
Cross-version (Gate 2) before a major tag: run compare-cross-version (reverse matrix, routes unset) on develop vs committed 6.0 baselines (baseline-6.0-win.csv / baseline-6.0-linux.csv).

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

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

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

Paste / regenerate

Product refresh and heavier tables:

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

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

Performance

Performance

Why this comparison is fair

How to read the tables

Contents

Measurement environment

Windows (GitHub-hosted windows-latest)

Linux (GitHub-hosted ubuntu-latest)

macOS (GitHub-hosted macos-15-intel)

Saturation control

Block A — H1 plain

Block B — H2 TLS→H1

Block C — H3→H1

Windows — Titanium vs nginx vs HAProxy vs Envoy vs YARP

Reverse

MITM (TWP only)

Linux — Titanium vs nginx vs HAProxy vs Envoy vs YARP

Reverse

MITM (TWP only)

macOS — Titanium vs nginx vs HAProxy vs Envoy vs YARP

Reverse

MITM (TWP only)

Editions (CLI / Plus / Intercept)

Cross-version (7.0 vs 6.0)

Heavier reverse workloads

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

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

Windows — POST 64 KiB request + 64 KiB response

Linux — POST 64 KiB request + 64 KiB response

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

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

Architecture-sensitive

Windows

Linux

TLS termination cost (H1 TLS → cleartext origin)

Windows

Linux

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

Tiered cadence

Arm shards (comparison groups)

Gate thresholds

Paste / regenerate

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Clone this wiki locally

Windows (GitHub-hosted `windows-latest`)

Linux (GitHub-hosted `ubuntu-latest`)

macOS (GitHub-hosted `macos-15-intel`)