feat(transport/cc): make slow start pluggable and add hystart++ by omansfeld · Pull Request #3278 · mozilla/neqo

omansfeld · 2025-12-16T14:04:53Z

fdc7597 introduces the scaffolding to make slow start configurable by introducing a SlowStart trait and using a generic parameter with that trait bound in the ClassicCongestionControl struct. It also rewrites the current slow start implementation in classic_slow_start.rs.

df81bdb makes the slow start algorithm configurable at run-time via the CLI/ConnectionParams, similar to how it works for Cubic/NewReno.

As a reminder: fdc7597..df81bdb had already had a review cycle before I started working on HyStart++ and made the PR a draft again.

41e8c04 implements HyStart++ as per RFC 9406 with extensive tests added in 86b496d.

With the latest commit ae92b97 I re-architected the SlowStart trait and the order in which slow start functionality is called from the congestion controller. Hystart functionality is unchanged between that commit and the former implementation, this just improves architecture and makes it more extensible for future work.

It now separates functionality into the correct order of operations for slow start (check for exit -> do the exit -> apply growth if not exiting -> do congestion avoidance if not in slow start anymore) which also makes it easier to add things like SUSS or rapid-start that accelerate slow start growth (but not touch the exit heuristics) in the future.

To maybe make review easier, the main parts are:

neqo-transport/src/cc/hystart.rs contains the HyStart++ implementation.
neqo-transport/src/cc/tests/hystart.rs contains tests for the HyStart++ implementation.
neqo-transport/src/cc/classic_slow_start.rs has the standard slow start implementation that was previously hardcoded into classic_cc.rs.
neqo-transport/src/cc/classic_cc.rs has the core on_packets_acked flow that calls everything else and the trait definition for SlowStart. There is also a small change to on_packet_sent for round-tracking.
... and then there is a lot of changing function calls in tests, scaffolding, etc.

I think it might make it the easiest to look at classic_cc.rs first and then follow the code flow from there into the HyStart++ implementation. I'm not sure it makes a lot of sense to look at commits one-by-one because of the re-architecture in the latest commit.

I also have some qlogs from validating the implementation against h3.speed.cloudflare.com where one can see HyStart++ working as intended. See this qvis screenshot where it exits at the same window size that is later converged on during congestion avoidance:

codecov · 2025-12-16T14:12:58Z

Codecov Report

❌ Patch coverage is 95.11278% with 13 lines in your changes missing coverage. Please review.
✅ Project coverage is 94.00%. Comparing base (70a0ba6) to head (b356f8c).
⚠️ Report is 6 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3278      +/-   ##
==========================================
- Coverage   94.12%   94.00%   -0.13%     
==========================================
  Files         125      131       +6     
  Lines       38339    38811     +472     
  Branches    38339    38811     +472     
==========================================
+ Hits        36086    36483     +397     
- Misses       1412     1481      +69     
- Partials      841      847       +6

Flag	Coverage Δ
freebsd	`93.03% <92.10%> (-0.12%)`	⬇️
linux	`94.10% <95.11%> (+<0.01%)`	⬆️
macos	`93.98% <95.11%> (-0.02%)`	⬇️
windows	`94.09% <95.11%> (-0.01%)`	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Components	Coverage Δ
neqo-common	`98.54% <ø> (ø)`
neqo-crypto	`86.90% <ø> (ø)`
neqo-http3	`93.91% <ø> (ø)`
neqo-qpack	`94.34% <ø> (ø)`
neqo-transport	`94.92% <95.11%> (-0.05%)`	⬇️
neqo-udp	`82.97% <ø> (ø)`
mtu	`86.61% <ø> (ø)`

Copilot

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

github-actions · 2025-12-16T14:29:12Z

Failed Interop Tests

QUIC Interop Runner, client vs. server, differences relative to main at b3338f9.

neqo-pr as client	neqo-pr as server
neqo-pr vs. aioquic: A 🚀C1 neqo-pr vs. go-x-net: A BP BA neqo-pr vs. haproxy: A BP BA neqo-pr vs. kwik: BP BA neqo-pr vs. linuxquic: A 🚀~~L1 C1~~ neqo-pr vs. lsquic: L1 C1 neqo-pr vs. msquic: A L1 ⚠️L2 C1 neqo-pr vs. mvfst: A ⚠️L1 C1 neqo-pr vs. nginx: A ⚠️L1 BP BA neqo-pr vs. ngtcp2: A CM neqo-pr vs. picoquic: A ⚠️L1 neqo-pr vs. quic-go: A L1 neqo-pr vs. quiche: A ⚠️L1 C1 BP BA neqo-pr vs. quinn: A L1 🚀C1 neqo-pr vs. s2n-quic: A ⚠️L1 C1 🚀BP BA CM neqo-pr vs. tquic: S A L1 BP BA neqo-pr vs. xquic: A 🚀~~L1 C1~~	aioquic vs. neqo-pr: ⚠️L1 CM go-x-net vs. neqo-pr: CM kwik vs. neqo-pr: BP BA CM msquic vs. neqo-pr: ⚠️L1 CM mvfst vs. neqo-pr: Z A L1 C1 CM openssl vs. neqo-pr: LR M A CM quic-go vs. neqo-pr: CM quiche vs. neqo-pr: ⚠️C1 CM quinn vs. neqo-pr: V2 CM s2n-quic vs. neqo-pr: CM tquic vs. neqo-pr: CM xquic vs. neqo-pr: M CM

All results

Succeeded Interop Tests

QUIC Interop Runner, client vs. server

neqo-pr as client

neqo-pr vs. aioquic: H DC LR C20 M S R Z 3 B U L1 L2 🚀C1 C2 6 V2 BP BA
neqo-pr vs. go-x-net: H DC LR M B U L2 C2 6
neqo-pr vs. haproxy: H DC LR C20 M S R Z 3 B U L1 L2 C1 C2 6 V2
neqo-pr vs. kwik: H DC LR C20 M S R Z 3 B U A L1 L2 C1 C2 6 V2
neqo-pr vs. linuxquic: H DC LR C20 M S R Z 3 B U E 🚀L1 L2 🚀C1 C2 6 V2 BP BA CM
neqo-pr vs. lsquic: H DC LR C20 M S R Z 3 B U E A L2 C2 6 V2 BP BA CM
neqo-pr vs. msquic: H DC LR C20 M S R Z B U ⚠️L2 C2 6 V2 BP BA
neqo-pr vs. mvfst: H DC LR M R Z 3 B U ⚠️L1 L2 C2 6 BP BA
neqo-pr vs. neqo: H DC LR C20 M S R Z 3 B U E A L1 L2 C1 C2 6 V2 BP BA CM
neqo-pr vs. nginx: H DC LR C20 M S R Z 3 B U ⚠️L1 L2 C1 C2 6
neqo-pr vs. ngtcp2: H DC LR C20 M S R Z 3 B U E L1 L2 C1 C2 6 V2 BP BA
neqo-pr vs. picoquic: H DC LR C20 M S R Z 3 B U E ⚠️L1 L2 C1 C2 6 V2 BP BA
neqo-pr vs. quic-go: H DC LR C20 M S R Z 3 B U L2 C1 C2 6 BP BA
neqo-pr vs. quiche: H DC LR C20 M S R Z 3 B U ⚠️L1 L2 C2 6
neqo-pr vs. quinn: H DC LR C20 M S R Z 3 B U E L2 🚀C1 C2 6 BP BA
neqo-pr vs. s2n-quic: H DC LR C20 M S R 3 B U E ⚠️L1 L2 C2 6 🚀BP
neqo-pr vs. tquic: H DC LR C20 M R Z 3 B U L2 C1 C2 6
neqo-pr vs. xquic: H DC LR C20 M S R Z 3 B U 🚀L1 L2 🚀C1 C2 6 BP BA

neqo-pr as server

aioquic vs. neqo-pr: H DC LR C20 M S R Z 3 B U A ⚠️L1 L2 C1 C2 6 V2 BP BA
chrome vs. neqo-pr: 3
go-x-net vs. neqo-pr: H DC LR M B U A L2 C2 6 BP BA
kwik vs. neqo-pr: H DC LR C20 M S R Z 3 B U A L1 L2 C1 C2 6 V2
linuxquic vs. neqo-pr: H DC LR C20 M S R Z 3 B U E A L1 L2 C1 C2 6 V2 BP 🚀BA CM
lsquic vs. neqo-pr: H DC LR C20 M S R Z 3 B E A L1 L2 C1 C2 6 V2 BP BA CM
msquic vs. neqo-pr: H DC LR C20 M S R Z B U A ⚠️L1 L2 C1 C2 6 V2 BP BA
mvfst vs. neqo-pr: H DC LR M 3 B L2 C2 6 BP BA
neqo vs. neqo-pr: H DC LR C20 M S R Z 3 B U E A L1 L2 C1 C2 6 V2 BP BA CM
ngtcp2 vs. neqo-pr: H DC LR C20 M S R Z 3 B U E A L1 L2 C1 C2 6 V2 BP BA CM
openssl vs. neqo-pr: H DC C20 S R 3 B L2 C2 6 BP BA
picoquic vs. neqo-pr: H DC LR C20 M S R Z 3 B U E A L1 L2 C1 C2 6 V2 BP BA CM
quic-go vs. neqo-pr: H DC LR C20 M S R Z 3 B U A L1 L2 C1 C2 6 BP BA
quiche vs. neqo-pr: H DC LR M S R Z 3 B A L1 L2 ⚠️C1 C2 6 BP BA
quinn vs. neqo-pr: H DC LR C20 M S R Z 3 B U E A L1 L2 C1 C2 6 BP BA
s2n-quic vs. neqo-pr: H DC LR M S R 3 B E A L1 L2 C1 C2 6 BP BA
tquic vs. neqo-pr: H DC LR M S R Z 3 B A L1 L2 C1 C2 6 BP BA
xquic vs. neqo-pr: H DC LR C20 S R Z 3 B U A L1 L2 C1 C2 6 BP BA

Unsupported Interop Tests

QUIC Interop Runner, client vs. server

neqo-pr as client

neqo-pr vs. aioquic: E CM
neqo-pr vs. go-x-net: C20 S R Z 3 E L1 C1 V2 CM
neqo-pr vs. haproxy: E CM
neqo-pr vs. kwik: E CM
neqo-pr vs. msquic: 3 E CM
neqo-pr vs. mvfst: C20 S E V2 CM
neqo-pr vs. nginx: E V2 CM
neqo-pr vs. picoquic: CM
neqo-pr vs. quic-go: E V2 CM
neqo-pr vs. quiche: E V2 CM
neqo-pr vs. quinn: V2 CM
neqo-pr vs. s2n-quic: Z V2
neqo-pr vs. tquic: E V2 CM
neqo-pr vs. xquic: E V2 CM

neqo-pr as server

aioquic vs. neqo-pr: E
chrome vs. neqo-pr: H DC LR C20 M S R Z B U E A L1 L2 C1 C2 6 V2 BP BA CM
go-x-net vs. neqo-pr: C20 S R Z 3 E L1 C1 V2
kwik vs. neqo-pr: E
lsquic vs. neqo-pr: U
msquic vs. neqo-pr: 3 E
mvfst vs. neqo-pr: C20 S R U E V2
openssl vs. neqo-pr: Z U E L1 C1 V2
quic-go vs. neqo-pr: E V2
quiche vs. neqo-pr: C20 U E V2
s2n-quic vs. neqo-pr: C20 Z U V2
tquic vs. neqo-pr: C20 U E V2
xquic vs. neqo-pr: E V2

Copilot

Pull request overview

Copilot reviewed 6 out of 6 changed files in this pull request and generated 2 comments.

mxinden

In favor of the design itself.

I suggest only merging here once we have more than one slow start algorithm. Each pull request should improve the state of main. Merging here now only adds a dead abstraction, i.e. SlowStart, only implemented once. (Similar to trait CongestionControl today.)

omansfeld · 2025-12-16T16:21:46Z

In favor of the design itself.

I suggest only merging here once we have more than one slow start algorithm. Each pull request should improve the state of main. Merging here now only adds a dead abstraction, i.e. SlowStart, only implemented once. (Similar to trait CongestionControl today.)

I very much get that point. That said, what would you say is the best point to move forward with this work then?

Having a bunch of dependent PRs seems like it could easily become a rebasing nightmare and one big PR seems like it could become very cumbersome to review. I would also very much like to have the along-the-way feedback of incremental PRs.

Should this be reviewed and merged onto a feature-branch that then eventually gets merged onto main?

omansfeld · 2025-12-19T20:41:49Z

I suggest only merging here once we have more than one slow start algorithm. Each pull request should improve the state of main. Merging here now only adds a dead abstraction, i.e. SlowStart, only implemented once. (Similar to trait CongestionControl today.)

As per @mxinden's comment I made this PR a draft again and will also add HyStart++ as part of it. I might still ask for review on single commits along the way to have some incremental feedback.

codspeed-hq · 2025-12-19T22:08:31Z

Merging this PR will degrade performance by 12.71%

⚡ 2 improved benchmarks
❌ 3 regressed benchmarks
✅ 46 untouched benchmarks

⚠️ Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	`BASE`	`HEAD`	Efficiency
❌	WallTime	`neqo-s2n`	510.5 ms	584.8 ms	-12.71%
❌	WallTime	`quiche-neqo`	345.8 ms	358.2 ms	-3.45%
⚡	WallTime	`msquic-msquic`	482.2 ms	459.3 ms	+4.98%
❌	WallTime	`neqo-neqo-newreno-nopacing`	280 ms	292.1 ms	-4.14%
⚡	WallTime	`neqo-neqo-cubic-nopacing`	283.4 ms	274.4 ms	+3.28%

_{Comparing omansfeld:pluggable_slow_start (b356f8c) with main (70a0ba6)}

omansfeld · 2025-12-23T12:19:13Z

I added another commit to make slow start configurable via connection parameters and the CLI. Would welcome another review once you have some time @mxinden!

Also a question, just to make sure I understood the parameter code structure correctly:

ConnectionParameters is what actually sets the cc parameters for the sender and is also what we would set via prefs from Firefox for e.g. experimentation in neqo_glue, correct?

neqo/neqo-transport/src/sender.rs

Lines 36 to 44 in adcd4bd

    
           Self { 
        
               cc: match conn_params.get_cc_algorithm() { 
        
                   CongestionControlAlgorithm::NewReno => { 
        
                       Box::new(ClassicCongestionControl::new(NewReno::default(), pmtud)) 
        
                   } 
        
                   CongestionControlAlgorithm::Cubic => { 
        
                       Box::new(ClassicCongestionControl::new(Cubic::default(), pmtud)) 
        
                   } 
        
               },

But QuicParameters is used to set CLI flags in which case it passes them on to ConnectionParams when a client/server is created (of course it also contains more than just connection parameters).

neqo/neqo-bin/src/lib.rs

Lines 234 to 243 in adcd4bd

    
           pub fn get(&self, alpn: &str) -> ConnectionParameters { 
        
               let mut params = ConnectionParameters::default() 
        
                   .max_streams(StreamType::BiDi, self.max_streams_bidi) 
        
                   .max_streams(StreamType::UniDi, self.max_streams_uni) 
        
                   .idle_timeout(Duration::from_secs(self.idle_timeout)) 
        
                   .initial_rtt(Duration::from_millis(self.initial_rtt_ms)) 
        
                   .cc_algorithm(self.congestion_control) 
        
                   .pacing(!self.no_pacing) 
        
                   .pmtud(!self.no_pmtud) 
        
                   .sni_slicing(!self.no_sni_slicing);

But from Firefox we directly set our ConnectionParameters, so any defaults we have in Neqo basically don't matter if the consumer is Firefox, correct?

mxinden · 2025-12-23T16:21:28Z

Also a question, just to make sure I understood the parameter code structure correctly:

ConnectionParameters is what actually sets the cc parameters for the sender and is also what we would set via prefs from Firefox for e.g. experimentation in neqo_glue, correct?

correct

But QuicParameters is used to set CLI flags in which case it passes them on to ConnectionParams when a client/server is created (of course it also contains more than just connection parameters).

correct

But from Firefox we directly set our ConnectionParameters, so any defaults we have in Neqo basically don't matter if the consumer is Firefox, correct?

neqo_glue does not use QuicParameters. neqo_glue does use ConnectionParameters::default.

mxinden

Changes here thus far look good to me.

omansfeld · 2026-02-20T16:58:07Z

The HyStart++ implementation is done now, so I'm opening up for review! I updated the main PR description to resemble the current state.

- there was a bug where we would exit css, but the rtt was still above the threshold for entering css, so we re-entered on the same ack - now we either are in css and can exit, or we are not in css and can enter, but not go back and forth in a single ack - this is in line with the RFC and with what other implementations are doing (msquic, linux, freeBSD) - this results in slightly more aggressive growth, since now we have non-css growth on the ack that exits css and only go back into css on the next ack if the css exit was truly spurious

mxinden

Thanks for the changes. I will take another look.

mxinden · 2026-03-05T19:50:32Z

After in person discussions Max and I decided to not restrict HyStart++ to initial slow start only for the following reasons:

* simplifies the code in `classic_cc::on_packets_acked`

* RFC is only stating a SHOULD, so we are not in direct conflict

* after persistent congestion the previously established `ssthresh` might still be too high, so there might be loss before it is reached, thus HyStart++ still makes sense

Can you document this decision in the code? Sorry in case it is already documented and I missed it.

- also fixing some nits

omansfeld · 2026-03-06T15:03:02Z

Updated again in 89a5169. Addressed nits from @mxinden, added comment about the non-initial slow start decision and added SlowStart::reset() to erase state after persistent congestion. Also added another test that asserts this.

mxinden

Ready to merge from my end.

I will take another look tomorrow with a fresh mind, but that doesn't need to block here.

github-actions · 2026-03-12T17:59:21Z

Failed Interop Tests

QUIC Interop Runner, client vs. server, differences relative to main at 70a0ba6.

neqo-pr as client	neqo-pr as server
neqo-pr vs. go-x-net: BP BA neqo-pr vs. haproxy: 🚀M BP BA neqo-pr vs. kwik: BP BA neqo-pr vs. lsquic: baseline result missing neqo-pr vs. msquic: A L1 C1 neqo-pr vs. mvfst: A L1 C1 BA neqo-pr vs. neqo: Z A neqo-pr vs. nginx: BP BA neqo-pr vs. ngtcp2: CM neqo-pr vs. picoquic: A ⚠️BP BA neqo-pr vs. quic-go: A neqo-pr vs. quiche: L1 BP BA neqo-pr vs. s2n-quic: 🚀BP BA CM neqo-pr vs. tquic: S BP BA neqo-pr vs. xquic: A 🚀C1	aioquic vs. neqo-pr: Z ⚠️L1 CM go-x-net vs. neqo-pr: CM kwik vs. neqo-pr: Z BP BA CM lsquic vs. neqo-pr: Z 🚀C1 ⚠️BP msquic vs. neqo-pr: Z ⚠️C1 CM mvfst vs. neqo-pr: Z A L1 C1 CM neqo vs. neqo-pr: Z A ⚠️BA openssl vs. neqo-pr: LR M A BA CM picoquic vs. neqo-pr: Z CM quic-go vs. neqo-pr: ⚠️Z CM quiche vs. neqo-pr: Z CM quinn vs. neqo-pr: Z V2 CM s2n-quic vs. neqo-pr: ⚠️B L1 CM tquic vs. neqo-pr: Z CM xquic vs. neqo-pr: M CM

All results

Succeeded Interop Tests

QUIC Interop Runner, client vs. server

neqo-pr as client

neqo-pr vs. aioquic: H DC LR C20 M S R Z 3 B U A L1 L2 C1 C2 6 V2 BP BA
neqo-pr vs. go-x-net: H DC LR M B U A L2 C2 6
neqo-pr vs. haproxy: H DC LR C20 🚀M S R Z 3 B U A L1 L2 C1 C2 6 V2
neqo-pr vs. kwik: H DC LR C20 M S R Z 3 B U A L1 L2 C1 C2 6 V2
neqo-pr vs. linuxquic: H DC LR C20 M S R Z 3 B U E A L1 L2 C1 C2 6 V2 BP BA CM
neqo-pr vs. msquic: H DC LR C20 M S R Z B U L2 C2 6 V2 BP BA
neqo-pr vs. mvfst: H DC LR M R Z 3 B U L2 C2 6 BP
neqo-pr vs. neqo: H DC LR C20 M S R 3 B U E L1 L2 C1 C2 6 V2 BP BA CM
neqo-pr vs. nginx: H DC LR C20 M S R Z 3 B U A L1 L2 C1 C2 6
neqo-pr vs. ngtcp2: H DC LR C20 M S R Z 3 B U E A L1 L2 C1 C2 6 V2 BP BA
neqo-pr vs. picoquic: H DC LR C20 M S R Z 3 B U E L1 L2 C1 C2 6 V2 ⚠️BP BA
neqo-pr vs. quic-go: H DC LR C20 M S R Z 3 B U L1 L2 C1 C2 6 BP BA
neqo-pr vs. quiche: H DC LR C20 M S R Z 3 B U A L2 C1 C2 6
neqo-pr vs. quinn: H DC LR C20 M S R Z 3 B U E A L1 L2 C1 C2 6 BP BA
neqo-pr vs. s2n-quic: H DC LR C20 M S R 3 B U E A L1 L2 C1 C2 6 🚀BP
neqo-pr vs. tquic: H DC LR C20 M R Z 3 B U A L1 L2 C1 C2 6
neqo-pr vs. xquic: H DC LR C20 M S R Z 3 B U L1 L2 🚀C1 C2 6 BP BA

neqo-pr as server

aioquic vs. neqo-pr: H DC LR C20 M S R 3 B U A ⚠️L1 L2 C1 C2 6 V2 BP BA
chrome vs. neqo-pr: 3
go-x-net vs. neqo-pr: H DC LR M B U A L2 C2 6 BP BA
kwik vs. neqo-pr: H DC LR C20 M S R 3 B U A L1 L2 C1 C2 6 V2
linuxquic vs. neqo-pr: H DC LR C20 M S R Z 3 B U E A L1 L2 C1 C2 6 V2 BP BA CM
lsquic vs. neqo-pr: H DC LR C20 M S R 3 B E A L1 L2 🚀C1 C2 6 V2 ⚠️BP BA CM
msquic vs. neqo-pr: H DC LR C20 M S R B U A L1 L2 ⚠️C1 C2 6 V2 BP BA
mvfst vs. neqo-pr: H DC LR M 3 B L2 C2 6 BP BA
neqo vs. neqo-pr: H DC LR C20 M S R 3 B U E L1 L2 C1 C2 6 V2 BP ⚠️BA CM
ngtcp2 vs. neqo-pr: H DC LR C20 M S R Z 3 B U E A L1 L2 C1 C2 6 V2 BP BA CM
openssl vs. neqo-pr: H DC C20 S R 3 B L2 C2 6 BP
picoquic vs. neqo-pr: H DC LR C20 M S R 3 B U E A L1 L2 C1 C2 6 V2 BP BA
quic-go vs. neqo-pr: H DC LR C20 M S R ⚠️Z 3 B U A L1 L2 C1 C2 6 BP BA
quiche vs. neqo-pr: H DC LR M S R 3 B A L1 L2 C1 C2 6 BP BA
quinn vs. neqo-pr: H DC LR C20 M S R 3 B U E A L1 L2 C1 C2 6 BP BA
s2n-quic vs. neqo-pr: H DC LR M S R 3 ⚠️B E A ⚠️L1 L2 C1 C2 6 BP BA
tquic vs. neqo-pr: H DC LR M S R 3 B A L1 L2 C1 C2 6 BP BA
xquic vs. neqo-pr: H DC LR C20 S R Z 3 B U A L1 L2 C1 C2 6 BP BA

Unsupported Interop Tests

QUIC Interop Runner, client vs. server

neqo-pr as client

neqo-pr vs. aioquic: E CM
neqo-pr vs. go-x-net: C20 S R Z 3 E L1 C1 V2 CM
neqo-pr vs. haproxy: E CM
neqo-pr vs. kwik: E CM
neqo-pr vs. msquic: 3 E CM
neqo-pr vs. mvfst: C20 S E V2 CM
neqo-pr vs. nginx: E V2 CM
neqo-pr vs. picoquic: CM
neqo-pr vs. quic-go: E V2 CM
neqo-pr vs. quiche: E V2 CM
neqo-pr vs. quinn: V2 CM
neqo-pr vs. s2n-quic: Z V2
neqo-pr vs. tquic: E V2 CM
neqo-pr vs. xquic: E V2 CM

neqo-pr as server

aioquic vs. neqo-pr: E
chrome vs. neqo-pr: H DC LR C20 M S R Z B U E A L1 L2 C1 C2 6 V2 BP BA CM
go-x-net vs. neqo-pr: C20 S R Z 3 E L1 C1 V2
kwik vs. neqo-pr: E
lsquic vs. neqo-pr: U
msquic vs. neqo-pr: 3 E
mvfst vs. neqo-pr: C20 S R U E V2
openssl vs. neqo-pr: Z U E L1 C1 V2
quic-go vs. neqo-pr: E V2
quiche vs. neqo-pr: C20 U E V2
s2n-quic vs. neqo-pr: C20 Z U V2
tquic vs. neqo-pr: C20 U E V2
xquic vs. neqo-pr: E V2

github-actions · 2026-03-12T18:01:56Z

Client/server transfer results

Performance differences relative to 70a0ba6.

Transfer of 33554432 bytes over loopback, min. 100 runs. All unit-less numbers are in milliseconds.

Client vs. server (params)	Mean ± σ	Min	Max	MiB/s ± σ	Δ `main`	Δ `main`
neqo-neqo-cubic-nopacing	96.7 ± 4.2	89.5	105.2	330.9 ± 7.6	💔 1.7	1.8%
neqo-s2n-cubic	219.4 ± 4.2	211.7	228.6	145.9 ± 7.6	💔 2.2	1.0%
s2n-neqo-cubic	173.2 ± 5.3	162.8	197.4	184.8 ± 6.0	💚 -2.1	-1.2%

Table above only shows statistically significant changes. See all results below.

All results

Transfer of 33554432 bytes over loopback, min. 100 runs. All unit-less numbers are in milliseconds.

Client vs. server (params)	Mean ± σ	Min	Max	MiB/s ± σ	Δ `main`	Δ `main`
google-google-nopacing	456.4 ± 4.8	450.8	477.3	70.1 ± 6.7
google-neqo-cubic	275.0 ± 4.4	268.6	292.7	116.3 ± 7.3	-0.9	-0.3%
msquic-msquic-nopacing	232.5 ± 103.5	138.7	578.9	137.7 ± 0.3
msquic-neqo-cubic	217.3 ± 78.2	150.1	581.7	147.2 ± 0.4	-14.9	-6.4%
neqo-google-cubic	754.5 ± 4.8	748.5	781.8	42.4 ± 6.7	-0.0	-0.0%
neqo-msquic-cubic	159.6 ± 4.1	153.6	167.6	200.5 ± 7.8	0.9	0.5%
neqo-neqo-cubic	98.2 ± 4.4	88.7	107.4	325.8 ± 7.3	1.0	1.0%
neqo-neqo-cubic-nopacing	96.7 ± 4.2	89.5	105.2	330.9 ± 7.6	💔 1.7	1.8%
neqo-neqo-newreno	97.9 ± 4.9	86.7	108.3	326.7 ± 6.5	-0.0	-0.0%
neqo-neqo-newreno-nopacing	95.5 ± 3.9	88.6	103.5	334.9 ± 8.2	0.3	0.3%
neqo-quiche-cubic	190.5 ± 3.8	185.6	202.5	167.9 ± 8.4	-0.8	-0.4%
neqo-s2n-cubic	219.4 ± 4.2	211.7	228.6	145.9 ± 7.6	💔 2.2	1.0%
quiche-neqo-cubic	182.8 ± 7.5	171.8	221.6	175.0 ± 4.3	1.3	0.7%
quiche-quiche-nopacing	143.3 ± 4.6	136.8	155.8	223.3 ± 7.0
s2n-neqo-cubic	173.2 ± 5.3	162.8	197.4	184.8 ± 6.0	💚 -2.1	-1.2%
s2n-s2n-nopacing	247.9 ± 24.9	231.5	346.4	129.1 ± 1.3

Download data for profiler.firefox.com or download performance comparison data.

github-actions · 2026-03-12T18:10:41Z

Benchmark results

Significant performance differences relative to 70a0ba6.

transfer/1-conn/1-100mb-resp (aka. Download)/mtu-1504: 💔 Performance has regressed by +1.7538%.

       time:   [201.36 ms 201.96 ms 202.67 ms]
       thrpt:  [493.42 MiB/s 495.15 MiB/s 496.63 MiB/s]
change:
       time:   [+1.3976% +1.7538% +2.1063] (p = 0.00 < 0.05)
       thrpt:  [-2.0629% -1.7236% -1.3784]
       Performance has regressed.
Found 5 outliers among 100 measurements (5.00%)
3 (3.00%) high mild
2 (2.00%) high severe

transfer/1-conn/1-100mb-req (aka. Upload)/mtu-1504: 💔 Performance has regressed by +1.9825%.

       time:   [205.70 ms 206.09 ms 206.52 ms]
       thrpt:  [484.22 MiB/s 485.22 MiB/s 486.14 MiB/s]
change:
       time:   [+1.6768% +1.9825% +2.2961] (p = 0.00 < 0.05)
       thrpt:  [-2.2445% -1.9439% -1.6492]
       Performance has regressed.
Found 3 outliers among 100 measurements (3.00%)
1 (1.00%) low mild
1 (1.00%) high mild
1 (1.00%) high severe

streams/walltime/1000-streams/each-1000-bytes: 💔 Performance has regressed by +4.0151%.

       time:   [46.323 ms 46.386 ms 46.459 ms]
       change: [+3.8356% +4.0151% +4.2065] (p = 0.00 < 0.05)
       Performance has regressed.
Found 2 outliers among 100 measurements (2.00%)
1 (1.00%) high mild
1 (1.00%) high severe

All results

transfer/1-conn/1-100mb-resp (aka. Download)/mtu-1504: 💔 Performance has regressed by +1.7538%.

       time:   [201.36 ms 201.96 ms 202.67 ms]
       thrpt:  [493.42 MiB/s 495.15 MiB/s 496.63 MiB/s]
change:
       time:   [+1.3976% +1.7538% +2.1063] (p = 0.00 < 0.05)
       thrpt:  [-2.0629% -1.7236% -1.3784]
       Performance has regressed.
Found 5 outliers among 100 measurements (5.00%)
3 (3.00%) high mild
2 (2.00%) high severe

transfer/1-conn/10_000-parallel-1b-resp (aka. RPS)/mtu-1504: Change within noise threshold.

       time:   [280.19 ms 282.00 ms 283.81 ms]
       thrpt:  [35.235 Kelem/s 35.462 Kelem/s 35.690 Kelem/s]
change:
       time:   [-1.9902% -1.0605% -0.1034] (p = 0.02 < 0.05)
       thrpt:  [+0.1035% +1.0719% +2.0306]
       Change within noise threshold.

transfer/1-conn/1-1b-resp (aka. HPS)/mtu-1504: No change in performance detected.

       time:   [38.585 ms 38.745 ms 38.923 ms]
       thrpt:  [25.692   B/s 25.810   B/s 25.917   B/s]
change:
       time:   [-0.3016% +0.2723% +0.8239] (p = 0.37 > 0.05)
       thrpt:  [-0.8171% -0.2715% +0.3025]
       No change in performance detected.
Found 10 outliers among 100 measurements (10.00%)
4 (4.00%) high mild
6 (6.00%) high severe

transfer/1-conn/1-100mb-req (aka. Upload)/mtu-1504: 💔 Performance has regressed by +1.9825%.

       time:   [205.70 ms 206.09 ms 206.52 ms]
       thrpt:  [484.22 MiB/s 485.22 MiB/s 486.14 MiB/s]
change:
       time:   [+1.6768% +1.9825% +2.2961] (p = 0.00 < 0.05)
       thrpt:  [-2.2445% -1.9439% -1.6492]
       Performance has regressed.
Found 3 outliers among 100 measurements (3.00%)
1 (1.00%) low mild
1 (1.00%) high mild
1 (1.00%) high severe

decode 4096 bytes, mask ff: No change in performance detected.

       time:   [5.5696 µs 5.5794 µs 5.5903 µs]
       change: [-0.7636% -0.1624% +0.3988] (p = 0.63 > 0.05)
       No change in performance detected.
Found 18 outliers among 100 measurements (18.00%)
9 (9.00%) high mild
9 (9.00%) high severe

decode 1048576 bytes, mask ff: No change in performance detected.

       time:   [1.4285 ms 1.4339 ms 1.4426 ms]
       change: [+0.0206% +0.4687% +1.1248] (p = 0.08 > 0.05)
       No change in performance detected.
Found 2 outliers among 100 measurements (2.00%)
1 (1.00%) high mild
1 (1.00%) high severe

decode 4096 bytes, mask 7f: No change in performance detected.

       time:   [8.4654 µs 8.4788 µs 8.4933 µs]
       change: [-0.3905% -0.0249% +0.2978] (p = 0.89 > 0.05)
       No change in performance detected.
Found 5 outliers among 100 measurements (5.00%)
4 (4.00%) high mild
1 (1.00%) high severe

decode 1048576 bytes, mask 7f: No change in performance detected.

       time:   [2.1760 ms 2.1837 ms 2.1962 ms]
       change: [-0.2251% +0.1621% +0.8014] (p = 0.62 > 0.05)
       No change in performance detected.
Found 1 outliers among 100 measurements (1.00%)
1 (1.00%) high severe

decode 4096 bytes, mask 3f: No change in performance detected.

       time:   [6.9147 µs 6.9251 µs 6.9363 µs]
       change: [-0.1618% +0.1666% +0.4920] (p = 0.33 > 0.05)
       No change in performance detected.
Found 5 outliers among 100 measurements (5.00%)
2 (2.00%) high mild
3 (3.00%) high severe

decode 1048576 bytes, mask 3f: No change in performance detected.

       time:   [1.7668 ms 1.7741 ms 1.7859 ms]
       change: [-0.0929% +0.3650% +1.0351] (p = 0.26 > 0.05)
       No change in performance detected.
Found 1 outliers among 100 measurements (1.00%)
1 (1.00%) high severe

streams/simulated/1-streams/each-1000-bytes: No change in performance detected.

       time:   [129.68 ms 129.68 ms 129.69 ms]
       thrpt:  [7.5302 KiB/s 7.5304 KiB/s 7.5306 KiB/s]
change:
       time:   [-0.0047% -0.0010% +0.0029] (p = 0.63 > 0.05)
       thrpt:  [-0.0029% +0.0010% +0.0047]
       No change in performance detected.
Found 1 outliers among 100 measurements (1.00%)
1 (1.00%) high mild

streams/simulated/1000-streams/each-1-bytes: Change within noise threshold.

       time:   [2.5364 s 2.5366 s 2.5369 s]
       thrpt:  [394.18   B/s 394.22   B/s 394.27   B/s]
change:
       time:   [+0.0013% +0.0174% +0.0334] (p = 0.04 < 0.05)
       thrpt:  [-0.0334% -0.0174% -0.0013]
       Change within noise threshold.

streams/simulated/1000-streams/each-1000-bytes: No change in performance detected.

       time:   [6.5817 s 6.5877 s 6.5946 s]
       thrpt:  [148.08 KiB/s 148.24 KiB/s 148.38 KiB/s]
change:
       time:   [-0.3384% -0.1509% +0.0312] (p = 0.12 > 0.05)
       thrpt:  [-0.0312% +0.1511% +0.3395]
       No change in performance detected.
Found 2 outliers among 100 measurements (2.00%)
2 (2.00%) high severe

streams/walltime/1-streams/each-1000-bytes: Change within noise threshold.

       time:   [587.68 µs 589.96 µs 592.79 µs]
       change: [+0.5995% +1.1178% +1.6935] (p = 0.00 < 0.05)
       Change within noise threshold.
Found 8 outliers among 100 measurements (8.00%)
1 (1.00%) high mild
7 (7.00%) high severe

streams/walltime/1000-streams/each-1-bytes: Change within noise threshold.

       time:   [12.406 ms 12.423 ms 12.441 ms]
       change: [-0.6380% -0.4222% -0.2021] (p = 0.00 < 0.05)
       Change within noise threshold.

streams/walltime/1000-streams/each-1000-bytes: 💔 Performance has regressed by +4.0151%.

       time:   [46.323 ms 46.386 ms 46.459 ms]
       change: [+3.8356% +4.0151% +4.2065] (p = 0.00 < 0.05)
       Performance has regressed.
Found 2 outliers among 100 measurements (2.00%)
1 (1.00%) high mild
1 (1.00%) high severe

coalesce_acked_from_zero 1+1 entries: No change in performance detected.

       time:   [92.552 ns 92.830 ns 93.104 ns]
       change: [-1.2295% -0.2166% +0.5790] (p = 0.69 > 0.05)
       No change in performance detected.
Found 10 outliers among 100 measurements (10.00%)
7 (7.00%) high mild
3 (3.00%) high severe

coalesce_acked_from_zero 3+1 entries: No change in performance detected.

       time:   [110.35 ns 110.68 ns 111.02 ns]
       change: [-3.4883% -1.0878% +0.3176] (p = 0.47 > 0.05)
       No change in performance detected.
Found 11 outliers among 100 measurements (11.00%)
2 (2.00%) high mild
9 (9.00%) high severe

coalesce_acked_from_zero 10+1 entries: No change in performance detected.

       time:   [109.93 ns 110.39 ns 110.91 ns]
       change: [-0.3048% +0.3136% +0.9892] (p = 0.36 > 0.05)
       No change in performance detected.
Found 12 outliers among 100 measurements (12.00%)
5 (5.00%) low mild
7 (7.00%) high severe

coalesce_acked_from_zero 1000+1 entries: No change in performance detected.

       time:   [94.568 ns 94.947 ns 95.587 ns]
       change: [-0.1114% +0.4768% +1.0125] (p = 0.11 > 0.05)
       No change in performance detected.
Found 10 outliers among 100 measurements (10.00%)
3 (3.00%) high mild
7 (7.00%) high severe

RxStreamOrderer::inbound_frame(): Change within noise threshold.

       time:   [108.09 ms 108.19 ms 108.29 ms]
       change: [-0.6652% -0.5509% -0.4337] (p = 0.00 < 0.05)
       Change within noise threshold.
Found 1 outliers among 100 measurements (1.00%)
1 (1.00%) high mild

sent::Packets::take_ranges: No change in performance detected.

       time:   [4.5181 µs 4.6285 µs 4.7448 µs]
       change: [-3.0163% +0.3531% +3.8585] (p = 0.85 > 0.05)
       No change in performance detected.
Found 3 outliers among 100 measurements (3.00%)
3 (3.00%) high mild

transfer/simulated/pacing-false/varying-seeds: No change in performance detected.

       time:   [23.941 s 23.941 s 23.941 s]
       thrpt:  [171.09 KiB/s 171.09 KiB/s 171.09 KiB/s]
change:
       time:   [+0.0000% +0.0000% +0.0000] (p = NaN > 0.05)
       thrpt:  [+0.0000% +0.0000% +0.0000]
       No change in performance detected.

transfer/simulated/pacing-true/varying-seeds: No change in performance detected.

       time:   [23.676 s 23.676 s 23.676 s]
       thrpt:  [173.01 KiB/s 173.01 KiB/s 173.01 KiB/s]
change:
       time:   [+0.0000% +0.0000% +0.0000] (p = NaN > 0.05)
       thrpt:  [+0.0000% +0.0000% +0.0000]
       No change in performance detected.

transfer/simulated/pacing-false/same-seed: No change in performance detected.

       time:   [23.941 s 23.941 s 23.941 s]
       thrpt:  [171.09 KiB/s 171.09 KiB/s 171.09 KiB/s]
change:
       time:   [+0.0000% +0.0000% +0.0000] (p = NaN > 0.05)
       thrpt:  [+0.0000% +0.0000% +0.0000]
       No change in performance detected.

transfer/simulated/pacing-true/same-seed: No change in performance detected.

       time:   [23.676 s 23.676 s 23.676 s]
       thrpt:  [173.01 KiB/s 173.01 KiB/s 173.01 KiB/s]
change:
       time:   [+0.0000% +0.0000% +0.0000] (p = NaN > 0.05)
       thrpt:  [+0.0000% +0.0000% +0.0000]
       No change in performance detected.

transfer/walltime/pacing-false/varying-seeds: Change within noise threshold.

       time:   [23.213 ms 23.251 ms 23.303 ms]
       change: [+1.1236% +1.3350% +1.5728] (p = 0.00 < 0.05)
       Change within noise threshold.
Found 3 outliers among 100 measurements (3.00%)
2 (2.00%) high mild
1 (1.00%) high severe

transfer/walltime/pacing-true/varying-seeds: Change within noise threshold.

       time:   [23.305 ms 23.326 ms 23.350 ms]
       change: [+0.4256% +0.5767% +0.7186] (p = 0.00 < 0.05)
       Change within noise threshold.
Found 1 outliers among 100 measurements (1.00%)
1 (1.00%) high severe

transfer/walltime/pacing-false/same-seed: Change within noise threshold.

       time:   [23.395 ms 23.418 ms 23.442 ms]
       change: [+0.9303% +1.0634% +1.1991] (p = 0.00 < 0.05)
       Change within noise threshold.
Found 2 outliers among 100 measurements (2.00%)
2 (2.00%) high mild

transfer/walltime/pacing-true/same-seed: Change within noise threshold.

       time:   [23.814 ms 23.843 ms 23.886 ms]
       change: [+0.7849% +0.9701% +1.1969] (p = 0.00 < 0.05)
       Change within noise threshold.
Found 3 outliers among 100 measurements (3.00%)
2 (2.00%) high mild
1 (1.00%) high severe

Download data for profiler.firefox.com or download performance comparison data.

omansfeld · 2026-03-13T11:03:10Z

@larseggert do you want to take another look or can I merge? The plan is to land this pref'd off in firefox with a new release next week and then run experiments on it (classic is still the default in neqo).

larseggert

LGTM. Minor comments.

Copilot AI review requested due to automatic review settings December 16, 2025 14:04

omansfeld requested review from KershawChang, larseggert, martinthomson and mxinden as code owners December 16, 2025 14:04

omansfeld temporarily deployed to claude-review December 16, 2025 14:04 — with GitHub Actions Inactive

omansfeld commented Dec 16, 2025

View reviewed changes

Comment thread neqo-transport/src/cc/tests/cubic.rs

omansfeld commented Dec 16, 2025

View reviewed changes

Comment thread neqo-transport/src/cc/tests/new_reno.rs

Copilot AI reviewed Dec 16, 2025

View reviewed changes

omansfeld requested a review from Copilot December 16, 2025 15:29

Copilot AI reviewed Dec 16, 2025

View reviewed changes

Comment thread neqo-transport/src/cc/classic_cc.rs Outdated

Comment thread neqo-transport/src/cc/classic_slow_start.rs Outdated

mxinden reviewed Dec 16, 2025

View reviewed changes

larseggert reviewed Dec 16, 2025

View reviewed changes

omansfeld changed the title ~~refactor(transport/cc): make slow start pluggable~~ feat(transport/cc): make slow start pluggable and add hystart++ Dec 19, 2025

omansfeld marked this pull request as draft December 19, 2025 20:39

omansfeld force-pushed the pluggable_slow_start branch 2 times, most recently from ba2537b to 36308ef Compare December 19, 2025 21:58

omansfeld requested a review from mxinden December 23, 2025 11:53

mxinden reviewed Dec 23, 2025

View reviewed changes

omansfeld force-pushed the pluggable_slow_start branch 2 times, most recently from beb57b7 to 86b496d Compare February 18, 2026 15:03

omansfeld marked this pull request as ready for review February 20, 2026 16:55

omansfeld added 2 commits March 5, 2026 19:51

fixing clippy nightly

6d0022b

omansfeld force-pushed the pluggable_slow_start branch from 4fde4f7 to 6d0022b Compare March 5, 2026 18:51

omansfeld temporarily deployed to claude-review March 5, 2026 18:51 — with GitHub Actions Inactive

mxinden reviewed Mar 5, 2026

View reviewed changes

Comment thread neqo-transport/src/cc/tests/mod.rs Outdated

use Cubic in test helper

3a803bc

omansfeld temporarily deployed to claude-review March 5, 2026 19:43 — with GitHub Actions Inactive

mxinden reviewed Mar 5, 2026

View reviewed changes

Comment thread neqo-transport/src/cc/classic_cc.rs

Comment thread neqo-transport/src/sender.rs Outdated

Comment thread neqo-transport/src/cc/tests/hystart.rs Outdated

add SlowStart::reset to erase state after persistent congestion

89a5169

- also fixing some nits

omansfeld temporarily deployed to claude-review March 6, 2026 14:52 — with GitHub Actions Inactive

omansfeld requested a review from mxinden March 9, 2026 12:10

This was referenced Mar 10, 2026

feat(stats): add a stat recording cubic's w_max #3431

Merged

feat(stats): add HyStart++ stats #3432

Merged

mxinden approved these changes Mar 11, 2026

View reviewed changes

Comment thread neqo-transport/src/cc/classic_cc.rs Outdated

omansfeld added 3 commits March 12, 2026 18:14

Merge branch 'main' into pluggable_slow_start

1162504

Merge branch 'main' into pluggable_slow_start

7045450

make self immutable in calc_cwnd_increase

b356f8c

larseggert approved these changes Mar 13, 2026

View reviewed changes

Comment thread neqo-transport/src/cc/classic_cc.rs

Comment thread neqo-transport/src/cc/classic_cc.rs

Comment thread neqo-transport/src/cc/hystart.rs

larseggert added this pull request to the merge queue Mar 13, 2026

github-merge-queue Bot removed this pull request from the merge queue due to failed status checks Mar 13, 2026

larseggert added this pull request to the merge queue Mar 13, 2026

Merged via the queue into mozilla:main with commit 6a953c2 Mar 13, 2026
204 of 206 checks passed

omansfeld mentioned this pull request Mar 16, 2026

chore: prepare v0.24.0 #3481

Merged

Conversation

omansfeld commented Dec 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

codecov Bot commented Dec 16, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

github-actions Bot commented Dec 16, 2025

Failed Interop Tests

Succeeded Interop Tests

neqo-pr as client

neqo-pr as server

Unsupported Interop Tests

neqo-pr as client

neqo-pr as server

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

Uh oh!

Uh oh!

mxinden left a comment

Choose a reason for hiding this comment

Uh oh!

omansfeld commented Dec 16, 2025

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

omansfeld commented Dec 19, 2025

Uh oh!

codspeed-hq Bot commented Dec 19, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Merging this PR will degrade performance by 12.71%

Performance Changes

Uh oh!

omansfeld commented Dec 23, 2025

Uh oh!

mxinden commented Dec 23, 2025

Uh oh!

mxinden left a comment

Choose a reason for hiding this comment

Uh oh!

omansfeld commented Feb 20, 2026

Uh oh!

mxinden left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

mxinden commented Mar 5, 2026

Uh oh!

omansfeld commented Mar 6, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

mxinden left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

github-actions Bot commented Mar 12, 2026

Failed Interop Tests

Succeeded Interop Tests

neqo-pr as client

neqo-pr as server

Unsupported Interop Tests

neqo-pr as client

neqo-pr as server

Uh oh!

github-actions Bot commented Mar 12, 2026

omansfeld commented Dec 16, 2025 •

edited

Loading

codecov Bot commented Dec 16, 2025 •

edited

Loading

codspeed-hq Bot commented Dec 19, 2025 •

edited

Loading

omansfeld commented Mar 6, 2026 •

edited

Loading