Performance

Why this comparison is fair

These tables are a same-harness, same-origin, same-runner-class reverse-proxy comparison — not a blog-post bake-off of mismatched labs.

Same load generator (dotnet-httpclient), same origin process, same warmup/measure (2s / 8s), same concurrency ramp (8, 16, 32, 64), median of 3 GitHub Actions repeats.
Every reverse arm is three OS processes (load generator + origin child + proxy child). 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 (4-core-class). Laptop High-perf and macos-latest are never mixed into these tables.
YARP is Yarp.ReverseProxy 2.3.0 with equivalent TLS/ALPN. nginx uses streaming knobs (keepalive 256, buffering off). Linux nginx (mainline + http_v3) is authoritative; Windows nginx has no QUIC — those cells are Not possible, not losses.
MITM is TWP-only (CONNECT + forged certs). nginx/YARP cannot MITM; Lite/Full tables are overhead vs TWP reverse, not vs peers. There are no MITM charts.
The product signal is TWP÷YARP (gated ≥ 0.85 reverse) and MITM÷Reverse (Lite ≥ 0.50, Full ≥ 0.50), not absolute RPS. Absolute RPS moves with runner heat; ratios are the claim.
Product fast paths (session-lite, skip-poll, compressed relay) are in-tree defaults for interception-off reverse — the same class of work YARP/nginx do. 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 (one key for all forged leaves). That is an intentional RPS tradeoff: compromise of that material affects every host minted by that manager. Reverse/YARP/nginx paths do not mint leaves.

Numbers below are Release measurements with RpsLoadProbe. FiddlerCore is not compared (commercial debugger license; not a throughput peer).

For pooling knobs and certificate first-visit tuning, see Performance and pooling. For the local cool A/B lab, laptop tables, and profiling notes, see Performance Profiling.

Measurement environment
Windows — Titanium vs nginx vs YARP
Linux — Titanium vs nginx vs YARP
- Tiny JSON reverse is nginx’s best case on Linux
- Why isn’t HTTP/3 > HTTP/2 > HTTP/1 in raw RPS?
macOS — Titanium vs nginx vs YARP
- Reverse
- MITM (TWP only)
Editions (CLI / Plus / Intercept)
Cross-version (7.0 vs 6.0)
Heavier reverse workloads
Other measurements
Raising limits on large hosts

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 (workflow_dispatch RPS saturation: warmup 2s / measure 8s; concurrency 8, 16, 32, 64; median of 3 repeats; --stop-on-slo-fail default on). Every --ramp arm is three OS processes (parent load generator + origin child + proxy child), except origin-direct arms (load gen + origin only). Prefer TWP÷YARP / TWP÷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.

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
Release / wiki	`compare-product`	~3–4h
Heavier tables	`compare-bodies` / `post` / `lossy` / `arch` / `bridges` / `tls-cost`	dispatch independently

See PERF-GATES.md.

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)
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`)
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)
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 @ 024bd68d — 34126812894. 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).

pwsh tools/RpsLoadProbe/run-rps.ps1 -Mode compare-saturation

Block A — H1 plain

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

Windows (windows-latest)

Arm	Generator	Sustain	Peak	% of origin-HttpClient
origin-direct	dotnet-httpclient	50,248 _{(54 MiB / 40.0% CPU)}	50,248 _{(54 MiB / 40.0% CPU)}	100.0%
origin-direct-bombardier	bombardier	39,040 _{(55 MiB / 24.0% CPU)}	39,040 _{(55 MiB / 24.0% CPU)}	77.7%
bare-reverse-http1	dotnet-httpclient	25,031 _{(58 MiB / 44.0% CPU)}	25,031 _{(58 MiB / 44.0% CPU)}	49.8%
nginx-reverse-http1	dotnet-httpclient	13,418 _{(123 MiB / 24.8% CPU)}	13,418 _{(123 MiB / 24.8% CPU)}	26.7%
yarp-reverse-http1	dotnet-httpclient	21,010 _{(89 MiB / 50.2% CPU)}	21,010 _{(89 MiB / 50.2% CPU)}	41.8%
twp-reverse-http1	dotnet-httpclient	🥇 24,500 _{(75 MiB / 47.9% CPU)}	24,500 _{(75 MiB / 47.9% CPU)}	48.8%

Linux (ubuntu-latest)

Arm	Generator	Sustain	Peak	% of origin-HttpClient
origin-direct	dotnet-httpclient	71,493 _{(79 MiB / 42.6% CPU)}	71,493 _{(79 MiB / 42.6% CPU)}	100.0%
origin-direct-bombardier	bombardier	42,014 _{(79 MiB / 34.9% CPU)}	42,014 _{(79 MiB / 34.9% CPU)}	58.8%
bare-reverse-http1	dotnet-httpclient	32,717 _{(66 MiB / 45.0% CPU)}	32,717 _{(66 MiB / 45.0% CPU)}	45.8%
nginx-reverse-http1	dotnet-httpclient	🥇 39,137 _{(73 MiB / 40.8% CPU)}	39,137 _{(73 MiB / 40.8% CPU)}	54.7%
yarp-reverse-http1	dotnet-httpclient	27,676 _{(114 MiB / 49.8% CPU)}	27,676 _{(114 MiB / 49.8% CPU)}	38.7%
twp-reverse-http1	dotnet-httpclient	32,666 _{(85 MiB / 50.5% CPU)}	32,666 _{(85 MiB / 50.5% CPU)}	45.7%

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

Block B — H2 TLS→H1

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

Windows (windows-latest)

Arm	Generator	Sustain	Peak	÷YARP	÷nginx
nginx-reverse-http2	dotnet-httpclient	7,862 _{(139 MiB / 24.7% CPU)}	7,862 _{(139 MiB / 24.7% CPU)}	0.27×	1.00×
yarp-reverse-http2	dotnet-httpclient	28,862 _{(94 MiB / 53.2% CPU)}	28,862 _{(94 MiB / 53.2% CPU)}	1.00×	3.67×
twp-reverse-http2-cleartext	dotnet-httpclient	🥇 33,703 _{(103 MiB / 54.3% CPU)}	33,703 _{(103 MiB / 54.3% CPU)}	1.17×	4.29×

Linux (ubuntu-latest)

Arm	Generator	Sustain	Peak	÷YARP	÷nginx
nginx-reverse-http2	dotnet-httpclient	14,871 _{(98 MiB / 19.3% CPU)}	14,871 _{(98 MiB / 19.3% CPU)}	0.51×	1.00×
yarp-reverse-http2	dotnet-httpclient	29,038 _{(122 MiB / 50.0% CPU)}	29,038 _{(122 MiB / 50.0% CPU)}	1.00×	1.95×
twp-reverse-http2-cleartext	dotnet-httpclient	🥇 34,081 _{(123 MiB / 52.5% CPU)}	34,081 _{(123 MiB / 52.5% CPU)}	1.17×	2.29×

Block C — H3→H1

Same layout as Block B. Requires QuicListener; nginx only with http_v3_module (Windows nginx has no QUIC).

Windows (windows-latest)

Arm	Generator	Sustain	Peak	÷YARP	÷nginx
nginx-reverse-http3-cleartext	dotnet-httpclient	Not possible (no QUIC)	Not possible (no QUIC)	—	—
yarp-reverse-http3-cleartext	dotnet-httpclient	🥇 14,836 _{(158 MiB / 52.2% CPU)}	14,836 _{(158 MiB / 52.2% CPU)}	1.00×	—
twp-reverse-http3-cleartext	dotnet-httpclient	14,310 _{(104 MiB / 44.9% CPU)}	14,310 _{(104 MiB / 44.9% CPU)}	0.96×	—

Linux (ubuntu-latest)

Arm	Generator	Sustain	Peak	÷YARP	÷nginx
nginx-reverse-http3-cleartext	dotnet-httpclient	0 _{(104 MiB / 22.6% CPU)}	15,166 _{(104 MiB / 22.6% CPU)}	0.81×	1.00×
yarp-reverse-http3-cleartext	dotnet-httpclient	18,684 _{(181 MiB / 50.3% CPU)}	18,684 _{(181 MiB / 50.3% CPU)}	1.00×	1.23×
twp-reverse-http3-cleartext	dotnet-httpclient	🥇 20,638 _{(143 MiB / 50.7% CPU)}	20,638 _{(143 MiB / 50.7% CPU)}	1.10×	1.36×

How to read the tables

Reverse = bare transparent fixed-forward (no TWP plugins / interception). nginx knobs match TWP/YARP streaming (keepalive 256, proxy_buffering off). MITM = TWP-only table on the same Client×Origin wires: Lite = no-op handlers (unchanged-lite finish reuses reverse compressed relay); Full = mutating handlers that append up to four unique headers per direction (RPS harness adds one; product uses MitmCompressedRelayHelper — no probe name in library code). Remove/replace/non-unique header growth and body mutation still force full decode/re-encode. nginx/YARP cannot MITM. HTTP/3 has no cleartext client (QUIC always encrypted).
Sustainable = last concurrency that still met error/latency SLOs. Peak = highest RPS in that ramp.
🥇 = best among TWP / nginx / YARP on Reverse rows (or saturation blocks): highest RPS; on an RPS tie, lower Memory (RSS) then lower CPU%. MITM is TWP-only. Lite÷Reverse / Full÷Reverse = TWP MITM lite or full sustain ÷ TWP Reverse sustain on the same Client×Origin from the same compare-product job.
Not possible = product cannot do that path. Not measured = path exists but no published number yet for that OS.
Product refresh: compare-product @ 024bd68d — 34126809918. Heavier/saturation/tls:

pwsh tools/RpsLoadProbe/run-rps.ps1 -Mode compare-product
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

Windows — Titanium vs nginx 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 @ 024bd68d — compare-product 34126809918. 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. Laptop High-perf / cool-paired numbers stay on the local lab.

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

Windows reverse

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
HTTP/1 · plain	HTTP/1 · plain	🥇 43148 _{(72 MiB / 48.4% CPU)}	🥇 43148 _{(72 MiB / 48.4% CPU)}	27494 _{(123 MiB / 24.8% CPU)}	27494 _{(123 MiB / 24.8% CPU)}	37946 _{(88 MiB / 50.5% CPU)}	37946 _{(88 MiB / 50.5% CPU)}
HTTP/1 · plain	HTTP/1 · TLS	🥇 36643 _{(90 MiB / 50.4% CPU)}	🥇 36643 _{(90 MiB / 50.4% CPU)}	16661 _{(133 MiB / 24.7% CPU)}	16661 _{(133 MiB / 24.7% CPU)}	33622 _{(94 MiB / 48.9% CPU)}	33622 _{(94 MiB / 48.9% CPU)}
HTTP/1 · plain	HTTP/2 · plain	🥇 62491 _{(116 MiB / 48.5% CPU)}	🥇 62491 _{(116 MiB / 48.5% CPU)}	Not possible	Not possible	55620 _{(93 MiB / 52.7% CPU)}	55620 _{(93 MiB / 52.7% CPU)}
HTTP/1 · plain	HTTP/2 · TLS	🥇 54245 _{(128 MiB / 44.3% CPU)}	🥇 54245 _{(128 MiB / 44.3% CPU)}	Not possible	Not possible	49838 _{(96 MiB / 49.2% CPU)}	49838 _{(96 MiB / 49.2% CPU)}
HTTP/1 · plain	HTTP/3 · QUIC	29032 _{(116 MiB / 48.6% CPU)}	29032 _{(116 MiB / 48.6% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	🥇 29598 _{(124 MiB / 50.9% CPU)}	🥇 29598 _{(124 MiB / 50.9% CPU)}
HTTP/1 · TLS	HTTP/1 · plain	🥇 36898 _{(88 MiB / 47% CPU)}	🥇 36898 _{(88 MiB / 47% CPU)}	17311 _{(140 MiB / 24.9% CPU)}	17311 _{(140 MiB / 24.9% CPU)}	32283 _{(102 MiB / 49.6% CPU)}	32283 _{(102 MiB / 49.6% CPU)}
HTTP/1 · TLS	HTTP/1 · TLS	🥇 32801 _{(85 MiB / 48.7% CPU)}	🥇 32801 _{(85 MiB / 48.7% CPU)}	13080 _{(142 MiB / 24.9% CPU)}	13080 _{(142 MiB / 24.9% CPU)}	29170 _{(105 MiB / 50.9% CPU)}	29170 _{(105 MiB / 50.9% CPU)}
HTTP/1 · TLS	HTTP/2 · plain	🥇 49296 _{(127 MiB / 45.9% CPU)}	🥇 49296 _{(127 MiB / 45.9% CPU)}	Not possible	Not possible	45621 _{(107 MiB / 50.2% CPU)}	45621 _{(107 MiB / 50.2% CPU)}
HTTP/1 · TLS	HTTP/2 · TLS	🥇 43714 _{(134 MiB / 45.7% CPU)}	🥇 43714 _{(134 MiB / 45.7% CPU)}	Not possible	Not possible	42064 _{(113 MiB / 49.1% CPU)}	42064 _{(113 MiB / 49.1% CPU)}
HTTP/1 · TLS	HTTP/3 · QUIC	25085 _{(120 MiB / 50.4% CPU)}	25085 _{(120 MiB / 50.4% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	🥇 25200 _{(138 MiB / 52% CPU)}	🥇 25200 _{(138 MiB / 52% CPU)}
HTTP/2 · plain	HTTP/1 · plain	🥇 55581 _{(97 MiB / 52.1% CPU)}	🥇 55581 _{(97 MiB / 52.1% CPU)}	Not possible	Not possible	53465 _{(87 MiB / 50.3% CPU)}	53465 _{(87 MiB / 50.3% CPU)}
HTTP/2 · plain	HTTP/1 · TLS	🥇 46471 _{(104 MiB / 51% CPU)}	🥇 46471 _{(104 MiB / 51% CPU)}	Not possible	Not possible	44832 _{(97 MiB / 50.6% CPU)}	44832 _{(97 MiB / 50.6% CPU)}
HTTP/2 · plain	HTTP/2 · plain	🥇 150625 _{(57 MiB / 29.1% CPU)}	🥇 150625 _{(57 MiB / 29.1% CPU)}	Not possible	Not possible	88874 _{(93 MiB / 50.2% CPU)}	88874 _{(93 MiB / 50.2% CPU)}
HTTP/2 · plain	HTTP/2 · TLS	🥇 122710 _{(62 MiB / 27.4% CPU)}	🥇 122710 _{(62 MiB / 27.4% CPU)}	Not possible	Not possible	77808 _{(101 MiB / 48.3% CPU)}	77808 _{(101 MiB / 48.3% CPU)}
HTTP/2 · plain	HTTP/3 · QUIC	🥇 42885 _{(141 MiB / 53.2% CPU)}	🥇 42885 _{(141 MiB / 53.2% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	41031 _{(135 MiB / 51.7% CPU)}	41031 _{(135 MiB / 51.7% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	🥇 54084 _{(108 MiB / 51.7% CPU)}	🥇 54084 _{(108 MiB / 51.7% CPU)}	15726 _{(139 MiB / 24.3% CPU)}	15726 _{(139 MiB / 24.3% CPU)}	47838 _{(97 MiB / 51.2% CPU)}	47838 _{(97 MiB / 51.2% CPU)}
HTTP/2 · TLS	HTTP/1 · TLS	🥇 47790 _{(103 MiB / 51.4% CPU)}	🥇 47790 _{(103 MiB / 51.4% CPU)}	11737 _{(143 MiB / 24.4% CPU)}	11737 _{(143 MiB / 24.4% CPU)}	41198 _{(95 MiB / 51.8% CPU)}	41198 _{(95 MiB / 51.8% CPU)}
HTTP/2 · TLS	HTTP/2 · plain	🥇 140904 _{(74 MiB / 31.1% CPU)}	🥇 140904 _{(74 MiB / 31.1% CPU)}	Not possible	Not possible	77150 _{(101 MiB / 53% CPU)}	77150 _{(101 MiB / 53% CPU)}
HTTP/2 · TLS	HTTP/2 · TLS	🥇 116819 _{(75 MiB / 30.3% CPU)}	🥇 116819 _{(75 MiB / 30.3% CPU)}	Not possible	Not possible	70121 _{(98 MiB / 48.3% CPU)}	70121 _{(98 MiB / 48.3% CPU)}
HTTP/2 · TLS	HTTP/3 · QUIC	🥇 41432 _{(140 MiB / 50.9% CPU)}	🥇 41432 _{(140 MiB / 50.9% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	37958 _{(134 MiB / 51.3% CPU)}	37958 _{(134 MiB / 51.3% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	🥇 26401 _{(116 MiB / 44.6% CPU)}	🥇 26401 _{(116 MiB / 44.6% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	25984 _{(149 MiB / 48.5% CPU)}	25984 _{(149 MiB / 48.5% CPU)}
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 23473 _{(126 MiB / 43.8% CPU)}	🥇 23473 _{(126 MiB / 43.8% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	21408 _{(155 MiB / 49.8% CPU)}	21408 _{(155 MiB / 49.8% CPU)}
HTTP/3 · QUIC	HTTP/2 · plain	🥇 46358 _{(147 MiB / 47.8% CPU)}	🥇 46358 _{(147 MiB / 47.8% CPU)}	Not possible (no H3 to H2)	Not possible (no H3 to H2)	34679 _{(164 MiB / 49.7% CPU)}	34679 _{(164 MiB / 49.7% CPU)}
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 42019 _{(161 MiB / 42.9% CPU)}	🥇 42019 _{(161 MiB / 42.9% CPU)}	Not possible (no H3 to H2)	Not possible (no H3 to H2)	35763 _{(163 MiB / 47.9% CPU)}	35763 _{(163 MiB / 47.9% CPU)}
HTTP/3 · QUIC	HTTP/3 · QUIC	🥇 22676 _{(128 MiB / 45.5% CPU)}	🥇 22676 _{(128 MiB / 45.5% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	18594 _{(181 MiB / 51.2% CPU)}	18594 _{(181 MiB / 51.2% CPU)}

MITM (TWP only)

Same Client×Origin wires with interception on (compare-product 34126809918). 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/YARP cannot MITM. Lite÷Reverse / Full÷Reverse vs bare reverse (same job). Completion gate: Lite ≥ 0.50× and Full ≥ 0.50× reverse sustain @ c=64 (median of 3 GHA runs); reverse TWP÷YARP ≥ 0.85× (no nginx 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×. Current product floors are Lite ≥ 0.50× and Full ≥ 0.50× reverse (not a historical “all MITM ≥ 0.70×” claim).

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. Same 0.50 / 0.50 floors apply to later GHA medians. H2 multiplexed Full is the tightest arm (often ~0.50–0.68×).

Client	Origin	Lite sustain	Full sustain	Lite÷Reverse	Full÷Reverse
HTTP/1 · plain	HTTP/1 · plain	42416 _{(78 MiB / 49.6% CPU)}	41642 _{(82 MiB / 50.7% CPU)}	0.98×	0.97×
HTTP/1 · plain	HTTP/1 · TLS	35496 _{(93 MiB / 50.8% CPU)}	35017 _{(94 MiB / 49.4% CPU)}	0.97×	0.96×
HTTP/1 · plain	HTTP/2 · plain	60830 _{(118 MiB / 50.1% CPU)}	59596 _{(121 MiB / 49.7% CPU)}	0.97×	0.95×
HTTP/1 · plain	HTTP/2 · TLS	53754 _{(124 MiB / 47% CPU)}	52767 _{(135 MiB / 48.7% CPU)}	0.99×	0.97×
HTTP/1 · plain	HTTP/3 · QUIC	28558 _{(112 MiB / 51.6% CPU)}	28107 _{(114 MiB / 51% CPU)}	0.98×	0.97×
HTTP/1 · TLS	HTTP/1 · plain	35649 _{(93 MiB / 49.9% CPU)}	35247 _{(94 MiB / 50% CPU)}	0.97×	0.96×
HTTP/1 · TLS	HTTP/1 · TLS	32815 _{(94 MiB / 48.2% CPU)}	31979 _{(93 MiB / 49.9% CPU)}	1×	0.97×
HTTP/1 · TLS	HTTP/2 · plain	48342 _{(118 MiB / 47.9% CPU)}	47289 _{(123 MiB / 50.1% CPU)}	0.98×	0.96×
HTTP/1 · TLS	HTTP/2 · TLS	43891 _{(134 MiB / 46.9% CPU)}	43067 _{(125 MiB / 45.9% CPU)}	1×	0.99×
HTTP/1 · TLS	HTTP/3 · QUIC	25139 _{(116 MiB / 51.7% CPU)}	24602 _{(116 MiB / 50% CPU)}	1×	0.98×
HTTP/2 · plain	HTTP/1 · plain	55126 _{(101 MiB / 51.5% CPU)}	53456 _{(97 MiB / 53.2% CPU)}	0.99×	0.96×
HTTP/2 · plain	HTTP/1 · TLS	47504 _{(103 MiB / 54.9% CPU)}	46365 _{(103 MiB / 53.5% CPU)}	1.02×	1×
HTTP/2 · plain	HTTP/2 · plain	99759 _{(70 MiB / 49.8% CPU)}	93161 _{(74 MiB / 48.8% CPU)}	0.66×	0.62×
HTTP/2 · plain	HTTP/2 · TLS	87259 _{(79 MiB / 45.4% CPU)}	82403 _{(77 MiB / 46.7% CPU)}	0.71×	0.67×
HTTP/2 · plain	HTTP/3 · QUIC	42663 _{(137 MiB / 52.5% CPU)}	41785 _{(134 MiB / 54.6% CPU)}	0.99×	0.97×
HTTP/2 · TLS	HTTP/1 · plain	53275 _{(108 MiB / 53.8% CPU)}	51033 _{(112 MiB / 53% CPU)}	0.99×	0.94×
HTTP/2 · TLS	HTTP/1 · TLS	46834 _{(106 MiB / 54% CPU)}	44869 _{(111 MiB / 51.2% CPU)}	0.98×	0.94×
HTTP/2 · TLS	HTTP/2 · plain	95259 _{(97 MiB / 49.4% CPU)}	89191 _{(89 MiB / 46.6% CPU)}	0.68×	0.63×
HTTP/2 · TLS	HTTP/2 · TLS	84908 _{(85 MiB / 44.9% CPU)}	79447 _{(95 MiB / 45.9% CPU)}	0.73×	0.68×
HTTP/2 · TLS	HTTP/3 · QUIC	41573 _{(144 MiB / 52.7% CPU)}	40421 _{(146 MiB / 52.7% CPU)}	1×	0.98×
HTTP/3 · QUIC	HTTP/1 · plain	24842 _{(114 MiB / 45.9% CPU)}	23923 _{(124 MiB / 46.3% CPU)}	0.94×	0.91×
HTTP/3 · QUIC	HTTP/1 · TLS	22186 _{(131 MiB / 45.5% CPU)}	21112 _{(117 MiB / 45.2% CPU)}	0.95×	0.9×
HTTP/3 · QUIC	HTTP/2 · plain	42900 _{(157 MiB / 51.1% CPU)}	41039 _{(146 MiB / 50.2% CPU)}	0.93×	0.89×
HTTP/3 · QUIC	HTTP/2 · TLS	40022 _{(158 MiB / 47.9% CPU)}	39065 _{(157 MiB / 48.6% CPU)}	0.95×	0.93×
HTTP/3 · QUIC	HTTP/3 · QUIC	20576 _{(129 MiB / 48.4% CPU)}	20055 _{(129 MiB / 50.6% CPU)}	0.91×	0.88×

Linux — Titanium vs nginx vs YARP

Reverse

Median of 3 repeats on ubuntu-latest (4 vCPU / 16 GiB). Bare reverse 5×5 @ 024bd68d — compare-product 34126809918. Warmup 2s / measure 8s; concurrency 8, 16, 32, 64. Linux nginx is the authoritative nginx baseline. nginx terminate peers use keepalive 256 + streaming buffers. The RPS workflow installs nginx.org mainline (http_v3_module) and libmsquic. Prefer ratios over absolute RPS.

Linux reverse

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
HTTP/1 · plain	HTTP/1 · plain	45442 _{(82 MiB / 50% CPU)}	45442 _{(82 MiB / 50% CPU)}	🥇 54602 _{(73 MiB / 39.5% CPU)}	🥇 54602 _{(73 MiB / 39.5% CPU)}	40514 _{(117 MiB / 49.4% CPU)}	40514 _{(117 MiB / 49.4% CPU)}
HTTP/1 · plain	HTTP/1 · TLS	36887 _{(106 MiB / 49.2% CPU)}	36887 _{(106 MiB / 49.2% CPU)}	🥇 43002 _{(90 MiB / 41.1% CPU)}	🥇 43002 _{(90 MiB / 41.1% CPU)}	32675 _{(137 MiB / 49% CPU)}	32675 _{(137 MiB / 49% CPU)}
HTTP/1 · plain	HTTP/2 · plain	🥇 58774 _{(138 MiB / 50.8% CPU)}	🥇 58774 _{(138 MiB / 50.8% CPU)}	Not possible	Not possible	52131 _{(123 MiB / 48.3% CPU)}	52131 _{(123 MiB / 48.3% CPU)}
HTTP/1 · plain	HTTP/2 · TLS	🥇 45682 _{(155 MiB / 49.4% CPU)}	🥇 45682 _{(155 MiB / 49.4% CPU)}	Not possible	Not possible	43753 _{(127 MiB / 47.3% CPU)}	43753 _{(127 MiB / 47.3% CPU)}
HTTP/1 · plain	HTTP/3 · QUIC	🥇 31312 _{(144 MiB / 54.6% CPU)}	🥇 31312 _{(144 MiB / 54.6% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	29231 _{(156 MiB / 48.4% CPU)}	29231 _{(156 MiB / 48.4% CPU)}
HTTP/1 · TLS	HTTP/1 · plain	34763 _{(109 MiB / 49.9% CPU)}	34763 _{(109 MiB / 49.9% CPU)}	🥇 42628 _{(100 MiB / 41.1% CPU)}	🥇 42628 _{(100 MiB / 41.1% CPU)}	31712 _{(132 MiB / 49.9% CPU)}	31712 _{(132 MiB / 49.9% CPU)}
HTTP/1 · TLS	HTTP/1 · TLS	30596 _{(107 MiB / 47.7% CPU)}	30596 _{(107 MiB / 47.7% CPU)}	🥇 35876 _{(104 MiB / 40.9% CPU)}	🥇 35876 _{(104 MiB / 40.9% CPU)}	27094 _{(137 MiB / 49.5% CPU)}	27094 _{(137 MiB / 49.5% CPU)}
HTTP/1 · TLS	HTTP/2 · plain	🥇 43784 _{(145 MiB / 50.1% CPU)}	🥇 43784 _{(145 MiB / 50.1% CPU)}	Not possible	Not possible	39595 _{(140 MiB / 49.2% CPU)}	39595 _{(140 MiB / 49.2% CPU)}
HTTP/1 · TLS	HTTP/2 · TLS	🥇 38412 _{(155 MiB / 48.6% CPU)}	🥇 38412 _{(155 MiB / 48.6% CPU)}	Not possible	Not possible	35014 _{(142 MiB / 48.2% CPU)}	35014 _{(142 MiB / 48.2% CPU)}
HTTP/1 · TLS	HTTP/3 · QUIC	🥇 27078 _{(154 MiB / 53.3% CPU)}	🥇 27078 _{(154 MiB / 53.3% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	24650 _{(167 MiB / 49.1% CPU)}	24650 _{(167 MiB / 49.1% CPU)}
HTTP/2 · plain	HTTP/1 · plain	🥇 50993 _{(117 MiB / 51.6% CPU)}	🥇 50993 _{(117 MiB / 51.6% CPU)}	Not possible	Not possible	49389 _{(114 MiB / 47.9% CPU)}	49389 _{(114 MiB / 47.9% CPU)}
HTTP/2 · plain	HTTP/1 · TLS	🥇 40178 _{(129 MiB / 50.5% CPU)}	🥇 40178 _{(129 MiB / 50.5% CPU)}	Not possible	Not possible	38512 _{(124 MiB / 48.8% CPU)}	38512 _{(124 MiB / 48.8% CPU)}
HTTP/2 · plain	HTTP/2 · plain	🥇 112467 _{(91 MiB / 38.3% CPU)}	🥇 112467 _{(91 MiB / 38.3% CPU)}	Not possible	Not possible	68404 _{(124 MiB / 46.3% CPU)}	68404 _{(124 MiB / 46.3% CPU)}
HTTP/2 · plain	HTTP/2 · TLS	🥇 86598 _{(91 MiB / 37% CPU)}	🥇 86598 _{(91 MiB / 37% CPU)}	Not possible	Not possible	55654 _{(126 MiB / 44.9% CPU)}	55654 _{(126 MiB / 44.9% CPU)}
HTTP/2 · plain	HTTP/3 · QUIC	34438 _{(156 MiB / 50.5% CPU)}	34438 _{(156 MiB / 50.5% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	🥇 34482 _{(157 MiB / 45.7% CPU)}	🥇 34482 _{(157 MiB / 45.7% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	🥇 47100 _{(121 MiB / 51.4% CPU)}	🥇 47100 _{(121 MiB / 51.4% CPU)}	21439 _{(99 MiB / 18.8% CPU)}	21439 _{(99 MiB / 18.8% CPU)}	40159 _{(122 MiB / 48.2% CPU)}	40159 _{(122 MiB / 48.2% CPU)}
HTTP/2 · TLS	HTTP/1 · TLS	🥇 38515 _{(120 MiB / 49.1% CPU)}	🥇 38515 _{(120 MiB / 49.1% CPU)}	18594 _{(109 MiB / 19.3% CPU)}	18594 _{(109 MiB / 19.3% CPU)}	34597 _{(129 MiB / 48.5% CPU)}	34597 _{(129 MiB / 48.5% CPU)}
HTTP/2 · TLS	HTTP/2 · plain	🥇 103058 _{(99 MiB / 38.9% CPU)}	🥇 103058 _{(99 MiB / 38.9% CPU)}	Not possible	Not possible	54026 _{(123 MiB / 46.3% CPU)}	54026 _{(123 MiB / 46.3% CPU)}
HTTP/2 · TLS	HTTP/2 · TLS	🥇 77878 _{(99 MiB / 35.6% CPU)}	🥇 77878 _{(99 MiB / 35.6% CPU)}	Not possible	Not possible	48036 _{(131 MiB / 45% CPU)}	48036 _{(131 MiB / 45% CPU)}
HTTP/2 · TLS	HTTP/3 · QUIC	🥇 32383 _{(166 MiB / 49.6% CPU)}	🥇 32383 _{(166 MiB / 49.6% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	30252 _{(163 MiB / 45.2% CPU)}	30252 _{(163 MiB / 45.2% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	🥇 27304 _{(150 MiB / 51.3% CPU)}	🥇 27304 _{(150 MiB / 51.3% CPU)}	0 _{(108 MiB / 21.3% CPU)}	22735 _{(108 MiB / 21.3% CPU)}	25272 _{(191 MiB / 49.3% CPU)}	25272 _{(191 MiB / 49.3% CPU)}
HTTP/3 · QUIC	HTTP/1 · TLS	🥇 21505 _{(165 MiB / 48.8% CPU)}	🥇 21505 _{(165 MiB / 48.8% CPU)}	0 _{(119 MiB / 22.4% CPU)}	17246 _{(119 MiB / 22.4% CPU)}	21129 _{(197 MiB / 49.1% CPU)}	21129 _{(197 MiB / 49.1% CPU)}
HTTP/3 · QUIC	HTTP/2 · plain	🥇 36800 _{(166 MiB / 54.4% CPU)}	🥇 36800 _{(166 MiB / 54.4% CPU)}	Not possible (no H3 to H2)	Not possible (no H3 to H2)	32383 _{(198 MiB / 47.8% CPU)}	32383 _{(198 MiB / 47.8% CPU)}
HTTP/3 · QUIC	HTTP/2 · TLS	🥇 33722 _{(166 MiB / 51.6% CPU)}	🥇 33722 _{(166 MiB / 51.6% CPU)}	Not possible (no H3 to H2)	Not possible (no H3 to H2)	30263 _{(203 MiB / 46.8% CPU)}	30263 _{(203 MiB / 46.8% CPU)}
HTTP/3 · QUIC	HTTP/3 · QUIC	🥇 25361 _{(165 MiB / 44.6% CPU)}	🥇 25361 _{(165 MiB / 44.6% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	21111 _{(209 MiB / 45.8% CPU)}	21111 _{(209 MiB / 45.8% CPU)}

Tiny JSON reverse is nginx’s best case on Linux

The tiny keep-alive GET (small JSON, no request body) is where a native C reverse proxy spends almost no time in user code. On Linux, nginx often leads HTTP/1 → HTTP/1 (plain and TLS) because that path is a handful of epoll + writev loops with keepalive 256 and buffering off — the same shape the harness uses for TWP and YARP.

That is not a TWP regression and not the product gate. The gated signal is TWP÷YARP ≥ 0.85 on every reverse wire. nginx Not possible cells (H2/H3 upstream, Windows QUIC) are omitted from charts rather than plotted as zero. Multiplexed H2→H2 is TWP’s best reverse case (session-lite + compressed relay); nginx cannot peer that wire.

Why isn’t HTTP/3 > HTTP/2 > HTTP/1 in raw RPS?

Raw RPS on this probe is not “newer protocol = faster.” HTTP/3 (QUIC/UDP, TLS 1.3, QPACK) does more CPU per tiny GET than HTTP/2-over-TCP or HTTP/1 keep-alive. H3 shines under loss, HOL blocking, and connection migration — see the lossy / high-RTT tables — not on a lossless 4-core loopback microbenchmark.

HTTP/2 multiplexed same-protocol reverse can beat HTTP/1 here because one TCP connection carries many streams and TWP’s compressed-relay path avoids per-stream session objects. HTTP/1 still wins some terminate-lite H1→H1 wires (especially vs nginx on Linux). Compare within a protocol pair, using TWP÷YARP, not a stacked H3 > H2 > H1 ranking.

MITM (TWP only)

Same Client×Origin wires with interception on (compare-product 34126809918). 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/YARP cannot MITM. Lite÷Reverse / Full÷Reverse vs bare reverse (same job). Completion gate: Lite ≥ 0.50× and Full ≥ 0.50× reverse sustain @ c=64 (median of 3 GHA runs); reverse TWP÷YARP ≥ 0.85× (no nginx 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×. Current product floors are Lite ≥ 0.50× and Full ≥ 0.50× reverse (not a historical “all MITM ≥ 0.70×” claim).

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. Same 0.50 / 0.50 floors apply to later GHA medians. H2 multiplexed Full is the tightest arm (often ~0.50–0.68×).

Client	Origin	Lite sustain	Full sustain	Lite÷Reverse	Full÷Reverse
HTTP/1 · plain	HTTP/1 · plain	44176 _{(94 MiB / 50.8% CPU)}	39967 _{(95 MiB / 50% CPU)}	0.97×	0.88×
HTTP/1 · plain	HTTP/1 · TLS	33446 _{(109 MiB / 50.5% CPU)}	33296 _{(111 MiB / 49.5% CPU)}	0.91×	0.9×
HTTP/1 · plain	HTTP/2 · plain	51914 _{(136 MiB / 53.8% CPU)}	50316 _{(126 MiB / 53.7% CPU)}	0.88×	0.86×
HTTP/1 · plain	HTTP/2 · TLS	45010 _{(147 MiB / 50.3% CPU)}	41890 _{(143 MiB / 50.5% CPU)}	0.99×	0.92×
HTTP/1 · plain	HTTP/3 · QUIC	29698 _{(144 MiB / 55.2% CPU)}	29205 _{(154 MiB / 53.7% CPU)}	0.95×	0.93×
HTTP/1 · TLS	HTTP/1 · plain	32973 _{(115 MiB / 50.4% CPU)}	30304 _{(115 MiB / 50.2% CPU)}	0.95×	0.87×
HTTP/1 · TLS	HTTP/1 · TLS	28337 _{(113 MiB / 49% CPU)}	28230 _{(112 MiB / 48.3% CPU)}	0.93×	0.92×
HTTP/1 · TLS	HTTP/2 · plain	40580 _{(144 MiB / 51.3% CPU)}	38701 _{(142 MiB / 51.1% CPU)}	0.93×	0.88×
HTTP/1 · TLS	HTTP/2 · TLS	35623 _{(165 MiB / 50% CPU)}	34976 _{(157 MiB / 49.8% CPU)}	0.93×	0.91×
HTTP/1 · TLS	HTTP/3 · QUIC	24740 _{(158 MiB / 53.1% CPU)}	25654 _{(162 MiB / 53.3% CPU)}	0.91×	0.95×
HTTP/2 · plain	HTTP/1 · plain	44972 _{(117 MiB / 53.5% CPU)}	45432 _{(124 MiB / 53.5% CPU)}	0.88×	0.89×
HTTP/2 · plain	HTTP/1 · TLS	38000 _{(130 MiB / 51.6% CPU)}	34677 _{(126 MiB / 51.2% CPU)}	0.95×	0.86×
HTTP/2 · plain	HTTP/2 · plain	70311 _{(114 MiB / 52.2% CPU)}	62514 _{(110 MiB / 50.1% CPU)}	0.63×	0.56×
HTTP/2 · plain	HTTP/2 · TLS	56973 _{(106 MiB / 47.7% CPU)}	53478 _{(106 MiB / 47.1% CPU)}	0.66×	0.62×
HTTP/2 · plain	HTTP/3 · QUIC	32982 _{(160 MiB / 50.7% CPU)}	31395 _{(150 MiB / 50.6% CPU)}	0.96×	0.91×
HTTP/2 · TLS	HTTP/1 · plain	42980 _{(120 MiB / 52.6% CPU)}	40591 _{(123 MiB / 52.8% CPU)}	0.91×	0.86×
HTTP/2 · TLS	HTTP/1 · TLS	37190 _{(131 MiB / 51.1% CPU)}	33561 _{(133 MiB / 50.4% CPU)}	0.97×	0.87×
HTTP/2 · TLS	HTTP/2 · plain	65213 _{(122 MiB / 49.8% CPU)}	59028 _{(122 MiB / 48.6% CPU)}	0.63×	0.57×
HTTP/2 · TLS	HTTP/2 · TLS	57395 _{(114 MiB / 45.9% CPU)}	52587 _{(114 MiB / 45.6% CPU)}	0.74×	0.68×
HTTP/2 · TLS	HTTP/3 · QUIC	32026 _{(166 MiB / 50.4% CPU)}	30999 _{(167 MiB / 51% CPU)}	0.99×	0.96×
HTTP/3 · QUIC	HTTP/1 · plain	27096 _{(153 MiB / 52.9% CPU)}	25506 _{(160 MiB / 52.9% CPU)}	0.99×	0.93×
HTTP/3 · QUIC	HTTP/1 · TLS	20462 _{(165 MiB / 49.3% CPU)}	22732 _{(161 MiB / 49.6% CPU)}	0.95×	1.06×
HTTP/3 · QUIC	HTTP/2 · plain	31264 _{(168 MiB / 55.8% CPU)}	34476 _{(173 MiB / 55.9% CPU)}	0.85×	0.94×
HTTP/3 · QUIC	HTTP/2 · TLS	27463 _{(166 MiB / 52.5% CPU)}	30491 _{(168 MiB / 52.8% CPU)}	0.81×	0.9×
HTTP/3 · QUIC	HTTP/3 · QUIC	20207 _{(161 MiB / 50.1% CPU)}	23463 _{(166 MiB / 50.6% CPU)}	0.8×	0.93×

macOS — Titanium vs nginx vs YARP

Numbers are filled by tools/RpsLoadProbe/apply-wiki-paste.ps1 after compare-product on macos-15-intel (placeholder headers match the Linux table shape).

Reverse

Median of 3 repeats on macos-15-intel (4-core / 14 GB). Bare reverse 5×5 @ 024bd68d — compare-product 34126809918. 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 libmsquic (+ DYLD_*), and YARP. Do not publish from macos-latest (3-core / 7 GB).

macOS reverse

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
HTTP/1 · plain	HTTP/1 · plain	🥇 15024 _{(84 MiB / 34.6% CPU)}	🥇 15024 _{(84 MiB / 34.6% CPU)}	7464 _{(51 MiB / 13.8% CPU)}	7464 _{(51 MiB / 13.8% CPU)}	11453 _{(105 MiB / 35.2% CPU)}	11453 _{(105 MiB / 35.2% CPU)}
HTTP/1 · plain	HTTP/1 · TLS	🥇 11517 _{(94 MiB / 34.2% CPU)}	🥇 11517 _{(94 MiB / 34.2% CPU)}	6366 _{(73 MiB / 20.3% CPU)}	6366 _{(73 MiB / 20.3% CPU)}	11162 _{(122 MiB / 38.4% CPU)}	11162 _{(122 MiB / 38.4% CPU)}
HTTP/1 · plain	HTTP/2 · plain	17744 _{(89 MiB / 36.4% CPU)}	17744 _{(89 MiB / 36.4% CPU)}	Not possible	Not possible	🥇 22656 _{(111 MiB / 33.3% CPU)}	🥇 22656 _{(111 MiB / 33.3% CPU)}
HTTP/1 · plain	HTTP/2 · TLS	14182 _{(105 MiB / 36.8% CPU)}	14182 _{(105 MiB / 36.8% CPU)}	Not possible	Not possible	🥇 18762 _{(112 MiB / 34.9% CPU)}	🥇 18762 _{(112 MiB / 34.9% CPU)}
HTTP/1 · plain	HTTP/3 · QUIC	5226 _{(91 MiB / 42.5% CPU)}	5226 _{(91 MiB / 42.5% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	🥇 5914 _{(114 MiB / 35.4% CPU)}	🥇 5914 _{(114 MiB / 35.4% CPU)}
HTTP/1 · TLS	HTTP/1 · plain	🥇 10608 _{(95 MiB / 33.2% CPU)}	🥇 10608 _{(95 MiB / 33.2% CPU)}	6819 _{(73 MiB / 13.4% CPU)}	6819 _{(73 MiB / 13.4% CPU)}	9148 _{(118 MiB / 34.2% CPU)}	9148 _{(118 MiB / 34.2% CPU)}
HTTP/1 · TLS	HTTP/1 · TLS	🥇 8936 _{(98 MiB / 34.5% CPU)}	🥇 8936 _{(98 MiB / 34.5% CPU)}	6814 _{(81 MiB / 19.5% CPU)}	6814 _{(81 MiB / 19.5% CPU)}	8755 _{(130 MiB / 35% CPU)}	8755 _{(130 MiB / 35% CPU)}
HTTP/1 · TLS	HTTP/2 · plain	14872 _{(97 MiB / 35.8% CPU)}	14872 _{(97 MiB / 35.8% CPU)}	Not possible	Not possible	🥇 16912 _{(116 MiB / 32.6% CPU)}	🥇 16912 _{(116 MiB / 32.6% CPU)}
HTTP/1 · TLS	HTTP/2 · TLS	13282 _{(151 MiB / 33.6% CPU)}	13282 _{(151 MiB / 33.6% CPU)}	Not possible	Not possible	🥇 13433 _{(119 MiB / 31.7% CPU)}	🥇 13433 _{(119 MiB / 31.7% CPU)}
HTTP/1 · TLS	HTTP/3 · QUIC	3511 _{(98 MiB / 47.2% CPU)}	3511 _{(98 MiB / 47.2% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	🥇 5856 _{(124 MiB / 34% CPU)}	🥇 5856 _{(124 MiB / 34% CPU)}
HTTP/2 · plain	HTTP/1 · plain	🥇 20764 _{(89 MiB / 39.2% CPU)}	🥇 20764 _{(89 MiB / 39.2% CPU)}	Not possible	Not possible	17011 _{(105 MiB / 38.1% CPU)}	17011 _{(105 MiB / 38.1% CPU)}
HTTP/2 · plain	HTTP/1 · TLS	13358 _{(98 MiB / 37.9% CPU)}	13358 _{(98 MiB / 37.9% CPU)}	Not possible	Not possible	🥇 14990 _{(117 MiB / 41.9% CPU)}	🥇 14990 _{(117 MiB / 41.9% CPU)}
HTTP/2 · plain	HTTP/2 · plain	🥇 44499 _{(74 MiB / 28.2% CPU)}	🥇 44499 _{(74 MiB / 28.2% CPU)}	Not possible	Not possible	24595 _{(106 MiB / 36.3% CPU)}	24595 _{(106 MiB / 36.3% CPU)}
HTTP/2 · plain	HTTP/2 · TLS	🥇 27865 _{(78 MiB / 28.9% CPU)}	🥇 27865 _{(78 MiB / 28.9% CPU)}	Not possible	Not possible	23088 _{(113 MiB / 36.2% CPU)}	23088 _{(113 MiB / 36.2% CPU)}
HTTP/2 · plain	HTTP/3 · QUIC	6007 _{(96 MiB / 44.8% CPU)}	6007 _{(96 MiB / 44.8% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	🥇 7284 _{(116 MiB / 33.9% CPU)}	🥇 7284 _{(116 MiB / 33.9% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	🥇 19696 _{(93 MiB / 40.1% CPU)}	🥇 19696 _{(93 MiB / 40.1% CPU)}	11601 _{(72 MiB / 14.7% CPU)}	11601 _{(72 MiB / 14.7% CPU)}	14579 _{(112 MiB / 39.1% CPU)}	14579 _{(112 MiB / 39.1% CPU)}
HTTP/2 · TLS	HTTP/1 · TLS	🥇 14310 _{(98 MiB / 40.2% CPU)}	🥇 14310 _{(98 MiB / 40.2% CPU)}	9839 _{(91 MiB / 14.9% CPU)}	9839 _{(91 MiB / 14.9% CPU)}	11766 _{(122 MiB / 41.5% CPU)}	11766 _{(122 MiB / 41.5% CPU)}
HTTP/2 · TLS	HTTP/2 · plain	🥇 43713 _{(82 MiB / 29.7% CPU)}	🥇 43713 _{(82 MiB / 29.7% CPU)}	Not possible	Not possible	25843 _{(113 MiB / 37.6% CPU)}	25843 _{(113 MiB / 37.6% CPU)}
HTTP/2 · TLS	HTTP/2 · TLS	🥇 31290 _{(85 MiB / 28% CPU)}	🥇 31290 _{(85 MiB / 28% CPU)}	Not possible	Not possible	28790 _{(117 MiB / 37.3% CPU)}	28790 _{(117 MiB / 37.3% CPU)}
HTTP/2 · TLS	HTTP/3 · QUIC	🥇 7116 _{(109 MiB / 43.4% CPU)}	🥇 7116 _{(109 MiB / 43.4% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	6614 _{(123 MiB / 36.6% CPU)}	6614 _{(123 MiB / 36.6% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	5422 _{(94 MiB / 39.6% CPU)}	5422 _{(94 MiB / 39.6% CPU)}	0 _{(63 MiB / 11.4% CPU)}	9594 _{(63 MiB / 11.4% CPU)}	🥇 7112 _{(166 MiB / 35.9% CPU)}	🥇 7112 _{(166 MiB / 35.9% CPU)}
HTTP/3 · QUIC	HTTP/1 · TLS	5401 _{(117 MiB / 40.8% CPU)}	5401 _{(117 MiB / 40.8% CPU)}	0 _{(66 MiB / 11.9% CPU)}	6561 _{(66 MiB / 11.9% CPU)}	🥇 5996 _{(175 MiB / 36.7% CPU)}	🥇 5996 _{(175 MiB / 36.7% CPU)}
HTTP/3 · QUIC	HTTP/2 · plain	7072 _{(98 MiB / 40.3% CPU)}	7072 _{(98 MiB / 40.3% CPU)}	Not possible (no H3 to H2)	Not possible (no H3 to H2)	🥇 8446 _{(172 MiB / 34% CPU)}	🥇 8446 _{(172 MiB / 34% CPU)}
HTTP/3 · QUIC	HTTP/2 · TLS	6213 _{(106 MiB / 39% CPU)}	6213 _{(106 MiB / 39% CPU)}	Not possible (no H3 to H2)	Not possible (no H3 to H2)	🥇 7921 _{(167 MiB / 34.9% CPU)}	🥇 7921 _{(167 MiB / 34.9% CPU)}
HTTP/3 · QUIC	HTTP/3 · QUIC	4167 _{(97 MiB / 38.6% CPU)}	4167 _{(97 MiB / 38.6% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	🥇 4703 _{(188 MiB / 30.7% CPU)}	🥇 4703 _{(188 MiB / 30.7% CPU)}

MITM (TWP only)

Same Client×Origin wires with interception on (compare-product 34126809918). 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/YARP cannot MITM. Lite÷Reverse / Full÷Reverse vs bare reverse (same job). Completion gate: Lite ≥ 0.50× and Full ≥ 0.50× reverse sustain @ c=64 (median of 3 GHA runs); reverse TWP÷YARP ≥ 0.85× (no nginx 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×. Current product floors are Lite ≥ 0.50× and Full ≥ 0.50× reverse (not a historical “all MITM ≥ 0.70×” claim).

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. Same 0.50 / 0.50 floors apply to later GHA medians. H2 multiplexed Full is the tightest arm (often ~0.50–0.68×).

Client	Origin	Lite sustain	Full sustain	Lite÷Reverse	Full÷Reverse
HTTP/1 · plain	HTTP/1 · plain	19900 _{(85 MiB / 35% CPU)}	16913 _{(86 MiB / 35.3% CPU)}	1.32×	1.13×
HTTP/1 · plain	HTTP/1 · TLS	16485 _{(95 MiB / 39.2% CPU)}	15930 _{(95 MiB / 38.2% CPU)}	1.43×	1.38×
HTTP/1 · plain	HTTP/2 · plain	21395 _{(94 MiB / 39.2% CPU)}	22098 _{(94 MiB / 38.6% CPU)}	1.21×	1.25×
HTTP/1 · plain	HTTP/2 · TLS	18305 _{(112 MiB / 38.2% CPU)}	17064 _{(108 MiB / 38.8% CPU)}	1.29×	1.2×
HTTP/1 · plain	HTTP/3 · QUIC	5206 _{(92 MiB / 45.1% CPU)}	4197 _{(92 MiB / 44.5% CPU)}	1×	0.8×
HTTP/1 · TLS	HTTP/1 · plain	13388 _{(97 MiB / 34.1% CPU)}	9514 _{(97 MiB / 33.5% CPU)}	1.26×	0.9×
HTTP/1 · TLS	HTTP/1 · TLS	11780 _{(99 MiB / 36.4% CPU)}	8823 _{(100 MiB / 35.4% CPU)}	1.32×	0.99×
HTTP/1 · TLS	HTTP/2 · plain	10530 _{(99 MiB / 36.1% CPU)}	10040 _{(103 MiB / 36% CPU)}	0.71×	0.68×
HTTP/1 · TLS	HTTP/2 · TLS	8486 _{(155 MiB / 32.1% CPU)}	7775 _{(169 MiB / 33.9% CPU)}	0.64×	0.59×
HTTP/1 · TLS	HTTP/3 · QUIC	3027 _{(105 MiB / 48.8% CPU)}	2898 _{(105 MiB / 48.4% CPU)}	0.86×	0.83×
HTTP/2 · plain	HTTP/1 · plain	16212 _{(91 MiB / 39.3% CPU)}	14686 _{(91 MiB / 39.6% CPU)}	0.78×	0.71×
HTTP/2 · plain	HTTP/1 · TLS	13272 _{(98 MiB / 41.8% CPU)}	11864 _{(98 MiB / 40.3% CPU)}	0.99×	0.89×
HTTP/2 · plain	HTTP/2 · plain	28429 _{(80 MiB / 35.7% CPU)}	23104 _{(78 MiB / 34.1% CPU)}	0.64×	0.52×
HTTP/2 · plain	HTTP/2 · TLS	23386 _{(83 MiB / 35.7% CPU)}	20906 _{(84 MiB / 35.3% CPU)}	0.84×	0.75×
HTTP/2 · plain	HTTP/3 · QUIC	4491 _{(96 MiB / 45.1% CPU)}	4985 _{(97 MiB / 45.2% CPU)}	0.75×	0.83×
HTTP/2 · TLS	HTTP/1 · plain	15975 _{(94 MiB / 40.1% CPU)}	15511 _{(92 MiB / 40.5% CPU)}	0.81×	0.79×
HTTP/2 · TLS	HTTP/1 · TLS	16070 _{(97 MiB / 41.1% CPU)}	12680 _{(98 MiB / 43.1% CPU)}	1.12×	0.89×
HTTP/2 · TLS	HTTP/2 · plain	25016 _{(90 MiB / 33.1% CPU)}	26369 _{(87 MiB / 35.7% CPU)}	0.57×	0.6×
HTTP/2 · TLS	HTTP/2 · TLS	19676 _{(90 MiB / 31.6% CPU)}	24635 _{(88 MiB / 34.8% CPU)}	0.63×	0.79×
HTTP/2 · TLS	HTTP/3 · QUIC	5379 _{(106 MiB / 43.2% CPU)}	6026 _{(106 MiB / 40.5% CPU)}	0.76×	0.85×
HTTP/3 · QUIC	HTTP/1 · plain	4430 _{(103 MiB / 41.2% CPU)}	4710 _{(103 MiB / 38.8% CPU)}	0.82×	0.87×
HTTP/3 · QUIC	HTTP/1 · TLS	4132 _{(116 MiB / 41.7% CPU)}	3975 _{(115 MiB / 41.4% CPU)}	0.77×	0.74×
HTTP/3 · QUIC	HTTP/2 · plain	5296 _{(100 MiB / 38.6% CPU)}	5368 _{(100 MiB / 40.1% CPU)}	0.75×	0.76×
HTTP/3 · QUIC	HTTP/2 · TLS	5082 _{(108 MiB / 37.2% CPU)}	5304 _{(109 MiB / 39.7% CPU)}	0.82×	0.85×
HTTP/3 · QUIC	HTTP/3 · QUIC	3482 _{(97 MiB / 36.4% CPU)}	3026 _{(96 MiB / 37.8% CPU)}	0.84×	0.73×

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).

Gates prioritize SLO survival under load and reasonable overhead vs baseline. Pre-origin Plus middleware (CIDR/WAF/JWT/rate-limit) runs on H1 terminate-lite without SessionEventArgs; JWT caches successful bearer validations. Thresholds — see PERF-GATES.md.

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

pwsh tools/RpsLoadProbe/run-rps.ps1 -Mode compare-editions
pwsh tools/RpsLoadProbe/validate-edition-gates.ps1 -CsvPath tools/RpsLoadProbe/results/rps-ramp-*.csv

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

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

Cross-version (7.0 vs 6.0)

Gate 2 prerequisite before the v7.0.0 tag: run compare-cross-version (reverse matrix, routes unset) on develop and compare against committed 6.0 baselines from GHA 33087088466 (tools/RpsLoadProbe/results/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

Median of 3 @ 0ef6d4dd. Source: Actions 33270571908. Sustain RPS @ c=64 (SLO-pass median). Absolute 7.0÷6.0 ≈ 0.96–1.00× on same-protocol arms; peer-norm ≈ 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×

validate-cross-version.ps1 passed on both OS CSVs (RSS floor 1.20; peer-norm also passes when current TWP÷YARP ≥ 0.90). MITM arms live in compare-product / compare-mitm, not this reverse-only matrix.

Heavier reverse workloads

Separate from the tiny-GET matrix. Same measurement environments. Modes: compare-bodies, compare-post, compare-lossy, compare-tls-cost, compare-arch in RpsLoadProbe. Dispatch each independently via workflow_dispatch (no need to re-run full compare-product). 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. compare-arch is the slow-consumer / early-response / duplex set. Laptop numbers are on Performance Local Lab; CI medians go in the tables below.

Lossy link = userspace shim (not kernel netem): TCP gets per-buffer delay + occasional whole-connection stalls (honest HOL for multiplexed H2); UDP gets per-datagram delay + drops (QUIC). compare-lossy publishes H1/H2/H3; H3 is where the protocol design is supposed to matter.

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

Median of 3 repeats on windows-latest @ 024bd68d. Source: Actions 34126816180 (compare-bodies). Warmup 2s / measure 8s. RPS cells include (MiB / CPU%) footprints.

Body	Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
64 KiB	HTTP/1 · TLS	HTTP/1 · plain	🥇 8,810 _{(113 MiB / 47.9% CPU)}	8,810 _{(113 MiB / 47.9% CPU)}	548 _{(139 MiB / 24.7% CPU)}	548 _{(139 MiB / 24.7% CPU)}	7,794 _{(135 MiB / 45.9% CPU)}	7,794 _{(135 MiB / 45.9% CPU)}
64 KiB	HTTP/2 · TLS	HTTP/1 · plain	🥇 7,762 _{(178 MiB / 42.6% CPU)}	7,762 _{(178 MiB / 42.6% CPU)}	555 _{(140 MiB / 24.8% CPU)}	555 _{(140 MiB / 24.8% CPU)}	6,825 _{(132 MiB / 49.6% CPU)}	6,825 _{(132 MiB / 49.6% CPU)}
64 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 4,020 _{(135 MiB / 38.1% CPU)}	4,020 _{(135 MiB / 38.1% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	3,784 _{(193 MiB / 48.1% CPU)}	3,784 _{(193 MiB / 48.1% CPU)}
256 KiB	HTTP/1 · TLS	HTTP/1 · plain	🥇 2,842 _{(130 MiB / 46.4% CPU)}	2,842 _{(130 MiB / 46.4% CPU)}	0 _{(141 MiB / 24.8% CPU)}	162 _{(141 MiB / 24.8% CPU)}	2,600 _{(134 MiB / 45.6% CPU)}	2,600 _{(134 MiB / 45.6% CPU)}
256 KiB	HTTP/2 · TLS	HTTP/1 · plain	🥇 2,541 _{(147 MiB / 40.6% CPU)}	2,541 _{(147 MiB / 40.6% CPU)}	0 _{(140 MiB / 24.8% CPU)}	148 _{(140 MiB / 24.8% CPU)}	1,911 _{(134 MiB / 43.7% CPU)}	1,911 _{(134 MiB / 43.7% CPU)}
256 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 1,156 _{(108 MiB / 39.7% CPU)}	1,156 _{(108 MiB / 39.7% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	1,077 _{(170 MiB / 44.0% CPU)}	1,077 _{(170 MiB / 44.0% 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 @ 024bd68d. Source: Actions 34126816180 (compare-bodies). Warmup 2s / measure 8s.

Body	Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
64 KiB	HTTP/1 · TLS	HTTP/1 · plain	🥇 7,909 _{(169 MiB / 44.8% CPU)}	7,909 _{(169 MiB / 44.8% CPU)}	5,528 _{(97 MiB / 51.9% CPU)}	5,528 _{(97 MiB / 51.9% CPU)}	6,447 _{(167 MiB / 48.4% CPU)}	6,447 _{(167 MiB / 48.4% CPU)}
64 KiB	HTTP/2 · TLS	HTTP/1 · plain	🥇 5,792 _{(235 MiB / 38.6% CPU)}	5,792 _{(235 MiB / 38.6% CPU)}	1,554 _{(98 MiB / 18.2% CPU)}	1,554 _{(98 MiB / 18.2% CPU)}	4,814 _{(158 MiB / 46.8% CPU)}	4,814 _{(158 MiB / 46.8% CPU)}
64 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 5,478 _{(181 MiB / 44.1% CPU)}	5,478 _{(181 MiB / 44.1% CPU)}	0 _{(124 MiB / 21.4% CPU)}	1,696 _{(124 MiB / 21.4% CPU)}	4,331 _{(221 MiB / 52.3% CPU)}	4,331 _{(221 MiB / 52.3% CPU)}
256 KiB	HTTP/1 · TLS	HTTP/1 · plain	🥇 2,688 _{(126 MiB / 37.2% CPU)}	2,688 _{(126 MiB / 37.2% CPU)}	1,717 _{(97 MiB / 54.0% CPU)}	1,717 _{(97 MiB / 54.0% CPU)}	2,139 _{(168 MiB / 45.7% CPU)}	2,139 _{(168 MiB / 45.7% CPU)}
256 KiB	HTTP/2 · TLS	HTTP/1 · plain	🥇 1,533 _{(183 MiB / 32.6% CPU)}	1,533 _{(183 MiB / 32.6% CPU)}	578 _{(98 MiB / 23.1% CPU)}	578 _{(98 MiB / 23.1% CPU)}	1,381 _{(161 MiB / 46.0% CPU)}	1,381 _{(161 MiB / 46.0% CPU)}
256 KiB	HTTP/3 · QUIC	HTTP/1 · plain	🥇 1,458 _{(148 MiB / 43.5% CPU)}	1,458 _{(148 MiB / 43.5% CPU)}	0 _{(119 MiB / 20.5% CPU)}	125 _{(119 MiB / 20.5% CPU)}	1,248 _{(219 MiB / 48.1% CPU)}	1,248 _{(219 MiB / 48.1% 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 @ 024bd68d. Source: Actions 34126819267 (compare-post).

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
HTTP/1 · TLS	HTTP/1 · plain	🥇 6,578 _{(102 MiB / 44.2% CPU)}	6,578 _{(102 MiB / 44.2% CPU)}	381 _{(139 MiB / 24.9% CPU)}	381 _{(139 MiB / 24.9% CPU)}	4,448 _{(133 MiB / 56.5% CPU)}	4,448 _{(133 MiB / 56.5% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	🥇 4,521 _{(178 MiB / 48.8% CPU)}	4,521 _{(178 MiB / 48.8% CPU)}	370 _{(142 MiB / 24.8% CPU)}	370 _{(142 MiB / 24.8% CPU)}	3,851 _{(134 MiB / 51.9% CPU)}	3,851 _{(134 MiB / 51.9% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	🥇 2,211 _{(177 MiB / 41.2% CPU)}	2,211 _{(177 MiB / 41.2% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	2,182 _{(215 MiB / 48.9% CPU)}	2,182 _{(215 MiB / 48.9% 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 @ 024bd68d. Source: Actions 34126819267 (compare-post).

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
HTTP/1 · TLS	HTTP/1 · plain	🥇 7,829 _{(131 MiB / 40.9% CPU)}	7,829 _{(131 MiB / 40.9% CPU)}	5,642 _{(101 MiB / 39.5% CPU)}	5,642 _{(101 MiB / 39.5% CPU)}	5,323 _{(168 MiB / 53.5% CPU)}	5,323 _{(168 MiB / 53.5% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	🥇 4,898 _{(233 MiB / 43.8% CPU)}	4,898 _{(233 MiB / 43.8% CPU)}	2,242 _{(115 MiB / 17.1% CPU)}	2,242 _{(115 MiB / 17.1% CPU)}	3,998 _{(169 MiB / 49.7% CPU)}	3,998 _{(169 MiB / 49.7% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	🥇 3,979 _{(237 MiB / 44.7% CPU)}	3,979 _{(237 MiB / 44.7% CPU)}	985 _{(114 MiB / 23.9% CPU)}	985 _{(114 MiB / 23.9% CPU)}	3,302 _{(261 MiB / 50.9% CPU)}	3,302 _{(261 MiB / 50.9% 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 @ 024bd68d — 34126822092 (compare-lossy).

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
HTTP/1 · TLS	HTTP/1 · plain	🥇 662 _{(106 MiB / 5.1% CPU)}	662 _{(106 MiB / 5.1% CPU)}	639 _{(138 MiB / 19.3% CPU)}	639 _{(138 MiB / 19.3% CPU)}	659 _{(111 MiB / 5.9% CPU)}	659 _{(111 MiB / 5.9% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	0 _{(123 MiB / 1.9% CPU)}	85 _{(123 MiB / 1.9% CPU)}	0 _{(140 MiB / 0.8% CPU)}	17 _{(140 MiB / 0.8% CPU)}	0 _{(99 MiB / 1.1% CPU)}	16 _{(99 MiB / 1.1% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	0 _{(66 MiB / 0.0% CPU)}	0 _{(66 MiB / 0.0% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	0 _{(79 MiB / 0.0% CPU)}	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 remeasure kept above).

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

Median of 3 repeats @ 024bd68d. Source: 34126822092 (compare-lossy; lossy H3 uses quic-http3).

Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
HTTP/1 · TLS	HTTP/1 · plain	1,203 _{(144 MiB / 13.2% CPU)}	1,203 _{(144 MiB / 13.2% CPU)}	🥇 1,208 _{(98 MiB / 11.3% CPU)}	1,208 _{(98 MiB / 11.3% CPU)}	1,196 _{(150 MiB / 16.5% CPU)}	1,196 _{(150 MiB / 16.5% CPU)}
HTTP/2 · TLS	HTTP/1 · plain	🥇 313 _{(170 MiB / 6.9% CPU)}	313 _{(170 MiB / 6.9% CPU)}	40 _{(97 MiB / 0.3% CPU)}	40 _{(97 MiB / 0.3% CPU)}	40 _{(127 MiB / 1.4% CPU)}	40 _{(127 MiB / 1.4% CPU)}
HTTP/3 · QUIC	HTTP/1 · plain	324 _{(149 MiB / 12.9% CPU)}	324 _{(149 MiB / 12.9% CPU)}	90 _{(107 MiB / 2.4% CPU)}	90 _{(107 MiB / 2.4% CPU)}	🥇 357 _{(182 MiB / 20.8% CPU)}	357 _{(182 MiB / 20.8% CPU)}

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

Architecture-sensitive

compare-arch isolates slow app readers, origin-early response, H2 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 @ 024bd68d (34126829362) (compare-arch). Slow consumer = 256 KiB GET, 16 KiB read + 8 ms sleep. Early response = 64 KiB POST, origin writes after 8 KiB. Duplex H2 = overlapping 64 KiB POST on H2 TLS↔H2 TLS. WebSocket = echo round-trips/sec.

compare-lossy (slow network) is already published above; it is not a slow app reader.

Windows

Scenario	Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
Slow consumer (256 KiB GET, throttled client read)	HTTP/1 · TLS	HTTP/1 · plain	🥇 248 _{(98 MiB / 4.0% CPU)}	248 _{(98 MiB / 4.0% CPU)}	208 _{(141 MiB / 24.8% CPU)}	208 _{(141 MiB / 24.8% CPU)}	248 _{(108 MiB / 4.0% CPU)}	248 _{(108 MiB / 4.0% CPU)}
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/1 · plain	248 _{(115 MiB / 4.1% CPU)}	248 _{(115 MiB / 4.1% CPU)}	171 _{(139 MiB / 24.8% CPU)}	171 _{(139 MiB / 24.8% CPU)}	🥇 249 _{(111 MiB / 6.2% CPU)}	249 _{(111 MiB / 6.2% CPU)}
Slow consumer (256 KiB GET, throttled client read)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 264 _{(94 MiB / 18.1% CPU)}	264 _{(94 MiB / 18.1% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	263 _{(158 MiB / 20.7% CPU)}	263 _{(158 MiB / 20.7% CPU)}
Early response (origin writes after first request chunk)	HTTP/1 · TLS	HTTP/1 · plain	🥇 6,099 _{(99 MiB / 43.0% CPU)}	6,099 _{(99 MiB / 43.0% CPU)}	376 _{(140 MiB / 24.8% CPU)}	376 _{(140 MiB / 24.8% CPU)}	4,216 _{(136 MiB / 51.4% CPU)}	4,216 _{(136 MiB / 51.4% CPU)}
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/1 · plain	🥇 4,285 _{(178 MiB / 35.9% CPU)}	4,285 _{(178 MiB / 35.9% CPU)}	0 _{(141 MiB / 24.6% CPU)}	284 _{(141 MiB / 24.6% CPU)}	2,821 _{(135 MiB / 40.1% CPU)}	2,821 _{(135 MiB / 40.1% CPU)}
Early response (origin writes after first request chunk)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 2,252 _{(148 MiB / 41.0% CPU)}	2,252 _{(148 MiB / 41.0% CPU)}	Not possible (no QUIC)	Not possible (no QUIC)	1,721 _{(197 MiB / 45.2% CPU)}	1,721 _{(197 MiB / 45.2% CPU)}
Duplex (both directions live)	HTTP/2 · TLS	HTTP/2 · TLS	0 _{(92 MiB / 0.1% CPU)}	0 _{(92 MiB / 0.1% CPU)}	Not possible	Not possible	0 _{(114 MiB / 0.1% CPU)}	0 _{(114 MiB / 0.1% CPU)}
Duplex (WebSocket / extended CONNECT)	HTTP/1 · TLS	HTTP/1 · plain	🥇 31,275 _{(98 MiB / 40.2% CPU)}	31,275 _{(98 MiB / 40.2% CPU)}	18,071 _{(141 MiB / 24.6% CPU)}	18,071 _{(141 MiB / 24.6% CPU)}	28,913 _{(88 MiB / 43.6% CPU)}	28,913 _{(88 MiB / 43.6% CPU)}

Linux

Scenario	Client	Origin	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
Slow consumer (256 KiB GET, throttled client read)	HTTP/1 · TLS	HTTP/1 · plain	🥇 465 _{(118 MiB / 10.0% CPU)}	465 _{(118 MiB / 10.0% CPU)}	406 _{(97 MiB / 10.4% CPU)}	406 _{(97 MiB / 10.4% CPU)}	418 _{(147 MiB / 14.9% CPU)}	418 _{(147 MiB / 14.9% CPU)}
Slow consumer (256 KiB GET, throttled client read)	HTTP/2 · TLS	HTTP/1 · plain	🥇 472 _{(141 MiB / 20.2% CPU)}	472 _{(141 MiB / 20.2% CPU)}	259 _{(97 MiB / 7.7% CPU)}	259 _{(97 MiB / 7.7% CPU)}	467 _{(146 MiB / 25.7% CPU)}	467 _{(146 MiB / 25.7% CPU)}
Slow consumer (256 KiB GET, throttled client read)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 471 _{(129 MiB / 35.7% CPU)}	471 _{(129 MiB / 35.7% CPU)}	0 _{(118 MiB / 21.4% CPU)}	193 _{(118 MiB / 21.4% CPU)}	466 _{(196 MiB / 40.7% CPU)}	466 _{(196 MiB / 40.7% CPU)}
Early response (origin writes after first request chunk)	HTTP/1 · TLS	HTTP/1 · plain	🥇 4,568 _{(135 MiB / 47.8% CPU)}	4,568 _{(135 MiB / 47.8% CPU)}	3,545 _{(98 MiB / 49.3% CPU)}	3,545 _{(98 MiB / 49.3% CPU)}	3,027 _{(176 MiB / 56.5% CPU)}	3,027 _{(176 MiB / 56.5% CPU)}
Early response (origin writes after first request chunk)	HTTP/2 · TLS	HTTP/1 · plain	🥇 3,317 _{(220 MiB / 46.1% CPU)}	3,317 _{(220 MiB / 46.1% CPU)}	0 _{(114 MiB / 23.6% CPU)}	1,340 _{(114 MiB / 23.6% CPU)}	2,151 _{(169 MiB / 48.0% CPU)}	2,151 _{(169 MiB / 48.0% CPU)}
Early response (origin writes after first request chunk)	HTTP/3 · QUIC	HTTP/1 · plain	🥇 2,823 _{(204 MiB / 45.1% CPU)}	2,823 _{(204 MiB / 45.1% CPU)}	0 _{(114 MiB / 24.6% CPU)}	431 _{(114 MiB / 24.6% CPU)}	2,073 _{(243 MiB / 48.3% CPU)}	2,073 _{(243 MiB / 48.3% CPU)}
Duplex (both directions live)	HTTP/2 · TLS	HTTP/2 · TLS	0 _{(101 MiB / 0.1% CPU)}	0 _{(101 MiB / 0.1% CPU)}	Not possible	Not possible	0 _{(150 MiB / 0.3% CPU)}	0 _{(150 MiB / 0.3% CPU)}
Duplex (WebSocket / extended CONNECT)	HTTP/1 · TLS	HTTP/1 · plain	29,049 _{(124 MiB / 44.2% CPU)}	29,049 _{(124 MiB / 44.2% CPU)}	🥇 33,342 _{(98 MiB / 36.2% CPU)}	33,342 _{(98 MiB / 36.2% CPU)}	26,992 _{(123 MiB / 44.1% CPU)}	26,992 _{(123 MiB / 44.1% CPU)}

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

TLS termination cost (H1 TLS → cleartext origin)

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

Windows

Median of 3 repeats on windows-latest @ 024bd68d. Source: Actions 34126826292 (compare-tls-cost). Absolute RPS on GHA swings hard; prefer TWP÷YARP.

Workload	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
Keep-alive · tiny GET	🥇 20,329 _{(84 MiB / 47.8% CPU)}	20,329 _{(84 MiB / 47.8% CPU)}	8,644 _{(138 MiB / 24.8% CPU)}	8,644 _{(138 MiB / 24.8% CPU)}	17,804 _{(102 MiB / 50.3% CPU)}	17,804 _{(102 MiB / 50.3% CPU)}
New-connection · tiny GET	🥇 712 _{(85 MiB / 9.2% CPU)}	712 _{(85 MiB / 9.2% CPU)}	0 _{(138 MiB / 24.0% CPU)}	240 _{(138 MiB / 24.0% CPU)}	706 _{(110 MiB / 11.1% CPU)}	706 _{(110 MiB / 11.1% CPU)}
Keep-alive · 256 KiB GET	🥇 2,839 _{(118 MiB / 44.5% CPU)}	2,839 _{(118 MiB / 44.5% CPU)}	0 _{(141 MiB / 24.6% CPU)}	156 _{(141 MiB / 24.6% CPU)}	2,609 _{(137 MiB / 46.5% CPU)}	2,609 _{(137 MiB / 46.5% CPU)}

Linux

Median of 3 repeats @ 024bd68d. Source: Actions 34126826292 (compare-tls-cost).

Workload	TWP sustain	TWP peak	nginx sustain	nginx peak	YARP sustain	YARP peak
Keep-alive · tiny GET	23,672 _{(105 MiB / 49.9% CPU)}	23,672 _{(105 MiB / 49.9% CPU)}	🥇 27,804 _{(98 MiB / 41.6% CPU)}	27,804 _{(98 MiB / 41.6% CPU)}	20,362 _{(130 MiB / 50.1% CPU)}	20,362 _{(130 MiB / 50.1% CPU)}
New-connection · tiny GET	974 _{(121 MiB / 47.1% CPU)}	974 _{(121 MiB / 47.1% CPU)}	🥇 1,007 _{(98 MiB / 44.3% CPU)}	1,007 _{(98 MiB / 44.3% CPU)}	964 _{(149 MiB / 46.2% CPU)}	964 _{(149 MiB / 46.2% CPU)}
Keep-alive · 256 KiB GET	🥇 2,722 _{(126 MiB / 37.0% CPU)}	2,722 _{(126 MiB / 37.0% CPU)}	1,728 _{(97 MiB / 53.4% CPU)}	1,728 _{(97 MiB / 53.4% CPU)}	2,152 _{(169 MiB / 45.9% CPU)}	2,152 _{(169 MiB / 45.9% 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.

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;
};

Performance

Performance

Why this comparison is fair

Contents

Measurement environment

Tiered cadence

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 YARP

Reverse

MITM (TWP only)

Linux — Titanium vs nginx vs YARP

Reverse

Tiny JSON reverse is nginx’s best case on Linux

Why isn’t HTTP/3 > HTTP/2 > HTTP/1 in raw RPS?

MITM (TWP only)

macOS — Titanium vs nginx 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

Other measurements

Raising limits on large hosts

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`)