fix(proto): cap PTO backoff at 2 seconds post-handshake by dignifiedquire · Pull Request #523 · n0-computer/noq

dignifiedquire · 2026-03-21T11:50:29Z

Description

Cap PTO backoff post-handshake to match picoquic's picoquic_current_retransmit_timer(). Without a cap, PTO grows exponentially without bound during connectivity gaps, making recovery take tens of seconds even after the link is restored.

Relates to #376

Implementation

Matches picoquic's timing.c:42-88:

Idle timeout bound (timing.c:59-63): cap PTO at idle_timeout / 16 so ~16 retransmits fit before the idle timer fires. Only when idle_timeout > 15s.
Hard cap at 2s (timing.c:77-85): matches PICOQUIC_LARGE_RETRANSMIT_TIMER. For satellite paths (RTT > 610ms), uses 1.5 * smoothed_rtt instead.
PTO base time: use max(last_ack_eliciting, now) to prevent scheduling in the past when the cap makes duration small relative to how long ago the last packet was sent.

Only applies post-handshake to avoid interfering with connection establishment.

References

RFC 9002 §6.2.1 — PTO exponential backoff (no upper bound specified)
picoquic picoquic_current_retransmit_timer() — reference implementation

github-actions · 2026-03-21T11:52:55Z

Documentation for this PR has been generated and is available at: https://n0-computer.github.io/noq/pr/523/docs/noq/

Last updated: 2026-03-30T12:08:07Z

github-actions · 2026-03-21T12:04:09Z

Performance Comparison Report

84a2d4398ffba83c893290b5ad495bc4f921fc57 - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5901.0 Mbps	7996.1 Mbps	-26.2%	97.0% / 98.6%
medium-concurrent	5620.4 Mbps	7957.9 Mbps	-29.4%	96.5% / 98.1%
medium-single	4384.6 Mbps	4749.2 Mbps	-7.7%	98.7% / 100.0%
small-concurrent	4016.9 Mbps	5001.7 Mbps	-19.7%	97.0% / 99.6%
small-single	3673.3 Mbps	4684.1 Mbps	-21.6%	96.8% / 98.6%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	3016.7 Mbps	3938.6 Mbps	-23.4%
lan	782.4 Mbps	802.3 Mbps	-2.5%
lossy	69.8 Mbps	55.9 Mbps	+25.0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 21.9% slower on average

---

e31a3050c5bc177d303fa86e6fcbb8fb21b42550 - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5357.5 Mbps	7962.5 Mbps	-32.7%	93.5% / 98.3%
medium-concurrent	5524.5 Mbps	7827.3 Mbps	-29.4%	94.7% / 105.0%
medium-single	3617.4 Mbps	4749.4 Mbps	-23.8%	94.7% / 106.0%
small-concurrent	3815.0 Mbps	5276.4 Mbps	-27.7%	99.4% / 162.0%
small-single	3431.1 Mbps	4846.9 Mbps	-29.2%	90.7% / 98.1%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	3030.7 Mbps	4092.6 Mbps	-25.9%
lan	782.4 Mbps	810.3 Mbps	-3.4%
lossy	69.8 Mbps	69.8 Mbps	~0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 28.0% slower on average

---

b536af97431b460cd53ee38aec9210ca5a64b014 - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5361.5 Mbps	7840.7 Mbps	-31.6%	92.4% / 97.5%
medium-concurrent	5511.1 Mbps	7685.0 Mbps	-28.3%	91.7% / 96.9%
medium-single	3873.2 Mbps	4190.1 Mbps	-7.6%	88.5% / 96.9%
small-concurrent	3795.3 Mbps	4889.1 Mbps	-22.4%	90.7% / 98.3%
small-single	3492.1 Mbps	4346.8 Mbps	-19.7%	89.3% / 97.3%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	3092.6 Mbps	4006.1 Mbps	-22.8%
lan	782.4 Mbps	810.3 Mbps	-3.4%
lossy	69.8 Mbps	69.8 Mbps	~0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 23.2% slower on average

---

cac0daac3903724f201ae0eb22ac922b19ef086a - artifacts

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	2931.6 Mbps	N/A	N/A
lan	782.4 Mbps	N/A	N/A
lossy	69.8 Mbps	N/A	N/A
wan	83.8 Mbps	N/A	N/A

---

f6cf1dd3e6bdc7fe99c252152e112cb46bbfe684 - artifacts

No results available

---

f6cf1dd3e6bdc7fe99c252152e112cb46bbfe684 - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5382.6 Mbps	7900.0 Mbps	-31.9%	98.4% / 188.0%
medium-concurrent	5409.1 Mbps	7508.9 Mbps	-28.0%	95.0% / 130.0%
medium-single	4022.2 Mbps	4749.0 Mbps	-15.3%	93.0% / 106.0%
small-concurrent	3846.9 Mbps	5412.8 Mbps	-28.9%	91.4% / 99.4%
small-single	3459.2 Mbps	4866.4 Mbps	-28.9%	89.9% / 97.8%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	3099.7 Mbps	4134.5 Mbps	-25.0%
lan	782.4 Mbps	810.4 Mbps	-3.4%
lossy	69.8 Mbps	55.9 Mbps	+25.0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 26.4% slower on average

---

a18eddeda9a478bed04adb95f31f3d61732cd555 - artifacts

No results available

---

7c42112deeb9344aecbd0d509b19878cd263e9e2 - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5612.9 Mbps	7918.3 Mbps	-29.1%	96.6% / 109.0%
medium-concurrent	5642.8 Mbps	7833.6 Mbps	-28.0%	95.7% / 108.0%
medium-single	3800.5 Mbps	4511.8 Mbps	-15.8%	90.7% / 98.2%
small-concurrent	3944.9 Mbps	5290.8 Mbps	-25.4%	94.4% / 111.0%
small-single	3599.6 Mbps	4746.1 Mbps	-24.2%	96.5% / 112.0%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	3135.2 Mbps	3925.3 Mbps	-20.1%
lan	782.4 Mbps	810.3 Mbps	-3.4%
lossy	69.9 Mbps	69.8 Mbps	~0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 24.2% slower on average

---

6d2bcad430707dc090f3806606b8390603e9d1bd - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5938.5 Mbps	7968.6 Mbps	-25.5%	96.7% / 107.0%
medium-concurrent	5469.3 Mbps	7641.2 Mbps	-28.4%	95.8% / 108.0%
medium-single	4119.3 Mbps	4189.5 Mbps	-1.7%	92.9% / 109.0%
small-concurrent	3893.6 Mbps	4874.6 Mbps	-20.1%	93.0% / 108.0%
small-single	3604.2 Mbps	4402.4 Mbps	-18.1%	89.5% / 97.4%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	3072.5 Mbps	3663.9 Mbps	-16.1%
lan	782.5 Mbps	796.5 Mbps	-1.8%
lossy	69.9 Mbps	55.9 Mbps	+25.0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 19.7% slower on average

---

5a4f32450f2c78f641bf5d04cd2024e904623649 - artifacts

No results available

---

83a158e039422cd4b6773e9dc50c41268c122e8f - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5375.3 Mbps	8180.3 Mbps	-34.3%	96.4% / 129.0%
medium-concurrent	5420.1 Mbps	7457.0 Mbps	-27.3%	91.9% / 97.3%
medium-single	3842.2 Mbps	4155.1 Mbps	-7.5%	96.7% / 127.0%
small-concurrent	3814.9 Mbps	4996.3 Mbps	-23.6%	93.7% / 102.0%
small-single	3392.7 Mbps	4053.9 Mbps	-16.3%	84.9% / 95.7%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	3034.9 Mbps	3967.0 Mbps	-23.5%
lan	782.4 Mbps	810.3 Mbps	-3.4%
lossy	69.8 Mbps	69.9 Mbps	~0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 23.6% slower on average

---

8793168f020876502c340e8b8755832e22d90fca - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5792.7 Mbps	7823.0 Mbps	-26.0%	95.9% / 105.0%
medium-concurrent	5287.1 Mbps	7970.3 Mbps	-33.7%	92.7% / 104.0%
medium-single	4157.5 Mbps	4737.5 Mbps	-12.2%	97.3% / 160.0%
small-concurrent	3907.9 Mbps	5121.1 Mbps	-23.7%	90.7% / 98.8%
small-single	3557.2 Mbps	4675.8 Mbps	-23.9%	94.7% / 108.0%

Summary

noq is 25.1% slower on average

---

2c616ad496abb80df45f403da7b3a1418f98e537 - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5342.6 Mbps	7896.9 Mbps	-32.3%	94.9% / 125.0%
medium-concurrent	5429.7 Mbps	7825.9 Mbps	-30.6%	94.8% / 126.0%
medium-single	3610.2 Mbps	4749.0 Mbps	-24.0%	92.0% / 102.0%
small-concurrent	3794.7 Mbps	5297.1 Mbps	-28.4%	96.1% / 129.0%
small-single	3551.7 Mbps	4858.7 Mbps	-26.9%	94.6% / 128.0%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	3092.4 Mbps	3603.9 Mbps	-14.2%
lan	782.5 Mbps	796.4 Mbps	-1.7%
lossy	69.8 Mbps	69.8 Mbps	~0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 26.8% slower on average

---

8d218842889c6e2e80112ec00ada63d135d598c4 - artifacts

No results available

---

e69ecfb31cdbd7bbea687d1a49f7812e044f47a3 - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5361.6 Mbps	7834.2 Mbps	-31.6%	93.8% / 101.0%
medium-concurrent	5415.7 Mbps	7864.2 Mbps	-31.1%	91.5% / 97.1%
medium-single	3784.3 Mbps	4561.3 Mbps	-17.0%	91.4% / 98.8%
small-concurrent	3896.6 Mbps	5071.6 Mbps	-23.2%	99.2% / 130.0%
small-single	3494.8 Mbps	4654.8 Mbps	-24.9%	86.6% / 96.3%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	2905.2 Mbps	N/A	N/A
lan	775.1 Mbps	N/A	N/A
lossy	69.8 Mbps	N/A	N/A
wan	83.8 Mbps	N/A	N/A

Summary

noq is 26.8% slower on average

---

4fbdccc243c1eb04877b63520949c4de64c754bd - artifacts

No results available

---

185a6e067ff98d8c6ca6d7ddb11f02551e62b9d9 - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5560.5 Mbps	8062.3 Mbps	-31.0%	91.2% / 96.7%
medium-concurrent	5351.9 Mbps	7856.4 Mbps	-31.9%	98.8% / 191.0%
medium-single	4618.5 Mbps	4749.6 Mbps	-2.8%	90.3% / 97.6%
small-concurrent	3789.4 Mbps	5372.8 Mbps	-29.5%	97.9% / 135.0%
small-single	3348.5 Mbps	4821.1 Mbps	-30.5%	93.0% / 102.0%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	3083.4 Mbps	N/A	N/A
lan	782.5 Mbps	N/A	N/A
lossy	69.8 Mbps	N/A	N/A
wan	83.8 Mbps	N/A	N/A

Summary

noq is 26.5% slower on average

---

ce171940ac91339fd7fb6b32053dfbe66e56ad52 - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5460.0 Mbps	7875.9 Mbps	-30.7%	93.3% / 98.1%
medium-concurrent	5576.5 Mbps	7920.7 Mbps	-29.6%	91.4% / 97.0%
medium-single	4120.6 Mbps	4596.9 Mbps	-10.4%	94.8% / 107.0%
small-concurrent	3860.0 Mbps	5224.9 Mbps	-26.1%	96.5% / 107.0%
small-single	3527.2 Mbps	4690.7 Mbps	-24.8%	93.6% / 108.0%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	2863.0 Mbps	3631.3 Mbps	-21.2%
lan	782.5 Mbps	796.4 Mbps	-1.7%
lossy	69.8 Mbps	55.9 Mbps	+25.0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 24.5% slower on average

---

a948f32bdc33af9806c3074c622cc0a230e48f19 - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	6193.8 Mbps	7899.3 Mbps	-21.6%	97.1% / 98.6%
medium-concurrent	5521.5 Mbps	7625.9 Mbps	-27.6%	96.5% / 98.1%
medium-single	3862.2 Mbps	4189.1 Mbps	-7.8%	96.6% / 98.6%
small-concurrent	4000.2 Mbps	5170.6 Mbps	-22.6%	97.0% / 99.5%
small-single	3707.0 Mbps	4458.9 Mbps	-16.9%	96.7% / 98.8%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	N/A	3926.3 Mbps	N/A
lan	N/A	798.8 Mbps	N/A
lossy	N/A	69.8 Mbps	N/A
wan	N/A	83.8 Mbps	N/A

Summary

noq is 20.6% slower on average

---

0c7dee4bc03990a3fb7dded05022bdab800f3993 - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5456.2 Mbps	7818.6 Mbps	-30.2%	94.8% / 104.0%
medium-concurrent	5467.4 Mbps	7966.8 Mbps	-31.4%	95.4% / 109.0%
medium-single	4140.2 Mbps	4594.7 Mbps	-9.9%	100.1% / 162.0%
small-concurrent	3956.7 Mbps	5217.5 Mbps	-24.2%	91.1% / 99.1%
small-single	3491.8 Mbps	4727.0 Mbps	-26.1%	95.3% / 109.0%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	2961.0 Mbps	N/A	N/A
lan	782.4 Mbps	N/A	N/A
lossy	69.8 Mbps	N/A	N/A
wan	83.8 Mbps	N/A	N/A

Summary

noq is 25.8% slower on average

---

536791ce61a59004a10c716d76f6eceb4023959f - artifacts

No results available

---

75629fb94db052d2c134eecdb84feb9fed10be42 - artifacts

No results available

---

21893a499bd682022b5acb87ddc0e08553bc8f2a - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5209.2 Mbps	7949.8 Mbps	-34.5%	91.3% / 97.4%
medium-concurrent	5417.3 Mbps	7850.7 Mbps	-31.0%	92.8% / 97.4%
medium-single	3580.0 Mbps	4689.8 Mbps	-23.7%	96.5% / 128.0%
small-concurrent	3846.4 Mbps	5192.9 Mbps	-25.9%	88.3% / 98.3%
small-single	3417.6 Mbps	4714.5 Mbps	-27.5%	92.9% / 102.0%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	3144.6 Mbps	3952.9 Mbps	-20.4%
lan	782.4 Mbps	810.3 Mbps	-3.4%
lossy	69.8 Mbps	69.8 Mbps	~0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 27.6% slower on average

---

84c9e1177531cbd77e64ac469c60e550ed1b7f1b - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5487.2 Mbps	8031.2 Mbps	-31.7%	94.3% / 105.0%
medium-concurrent	5377.2 Mbps	7486.0 Mbps	-28.2%	92.2% / 97.1%
medium-single	4067.9 Mbps	4749.2 Mbps	-14.3%	86.6% / 95.8%
small-concurrent	3816.5 Mbps	5369.0 Mbps	-28.9%	94.0% / 103.0%
small-single	3434.3 Mbps	4852.6 Mbps	-29.2%	89.9% / 97.4%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	3065.8 Mbps	3897.5 Mbps	-21.3%
lan	782.4 Mbps	805.9 Mbps	-2.9%
lossy	69.8 Mbps	55.9 Mbps	+25.0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 25.9% slower on average

---

b24e9ef4011b0221e13f9f8edfa0aa5171f58409 - artifacts

No results available

---

24b4c76a40f16238faee0ff0b82047740cf32d6b - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5669.7 Mbps	7928.0 Mbps	-28.5%	92.1% / 97.3%
medium-concurrent	5496.3 Mbps	7404.3 Mbps	-25.8%	98.7% / 181.0%
medium-single	4438.0 Mbps	4760.8 Mbps	-6.8%	95.9% / 122.0%
small-concurrent	3840.3 Mbps	5341.5 Mbps	-28.1%	95.9% / 125.0%
small-single	3503.6 Mbps	4713.2 Mbps	-25.7%	92.4% / 102.0%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	2877.3 Mbps	3925.3 Mbps	-26.7%
lan	782.1 Mbps	800.8 Mbps	-2.3%
lossy	69.9 Mbps	55.9 Mbps	+25.0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 23.6% slower on average

---

d1566b5115c9c14868029188dd8c9a8ff1e0cdf1 - artifacts

Raw Benchmarks (localhost)

Scenario	noq	upstream	Delta	CPU (avg/max)
large-single	5807.2 Mbps	8064.7 Mbps	-28.0%	92.6% / 97.7%
medium-concurrent	5290.2 Mbps	7795.9 Mbps	-32.1%	93.7% / 105.0%
medium-single	3798.8 Mbps	4189.7 Mbps	-9.3%	95.8% / 108.0%
small-concurrent	3883.3 Mbps	5055.1 Mbps	-23.2%	96.4% / 111.0%
small-single	3527.9 Mbps	4467.7 Mbps	-21.0%	93.1% / 108.0%

Netsim Benchmarks (network simulation)

Condition	noq	upstream	Delta
ideal	3037.5 Mbps	3939.2 Mbps	-22.9%
lan	782.5 Mbps	810.3 Mbps	-3.4%
lossy	69.8 Mbps	55.9 Mbps	+25.0%
wan	83.8 Mbps	83.8 Mbps	~0%

Summary

noq is 23.7% slower on average

divagant-martian · 2026-03-23T01:43:13Z

checked locally and the failing tests just do not finish

dignifiedquire · 2026-03-23T10:26:53Z

should be better now

#527 is being reverted upstream (#534). The busy-loop was caused by the PTO cap bug (#523), not swallowed ENETUNREACH. Proper fix needs a notify_destination_unreachable API in noq-proto (deferred).

flub

Generally LGTM though I do have some questions.

divagant-martian

I think I understand what the PR is aiming to solve, but I'm failing to understand the solution or why the approach is appropriate, beyond using picoquic as a sort of authoritative source.

Is the issue we are solving that PTO might become larger than the idle timeout? This is what I understood. I guess there are some other issues, but I'll focus on this one for the moment.

If this is the case, why divide by 16 only after 15? This makes for ~1s PTOs for idle timeouts of 15s but 2s PTOs for idle timeouts of 14s. It's such a weird criterion and I'm finding it hard to justify.

Why not cap the "desired" PTO to the min of something like 50% of the idle timeout and 2s, ensuring one PTO can always fire before the idle timeout (regardless of value)? Basically, I'm having issues with the only if idle_timeout > 15s. I've intentionally left out the hard cap special case of large RTT paths, which I think is a fine upper bound instead of the 2s.

I went ahead and gave a try at swapping picoquic's weird criterion with the one I propose, and all tests, including your new ones, pass. I'm keen on understanding if there are objections to this, but for now I'm inclined toward an approach that seems to work, is easier to understand, and does not have an arbitrary application criterion.

matheus23

There is another bug, that was hiding here, that his uncovers. Trying to fix it.. but we should keep this workaround in this PR imho

I disagree with you three.
There is definitely a bug in here. Why does noq-proto keep resending packets forever and never hit the idle timeout? This should not happen.
I'm looking at logs and will try to figure out the bug that this PR introduces.

For some reason I had a hard time understanding y'all here.
I'm fine with keeping the workaround in this PR and fixing the underlying bug ASAP.

The danger is that together with this PR's changes to the loss timer, there will be situations in which users will have broken connections that never time out. That's bad.

But I do see how the underlying bug for this is also present on main.

Perhaps we should fix it on main first, then remove the workaround here, then merge this?

dignifiedquire · 2026-03-25T10:29:56Z

I swapped this for now + duration and all tests work, am I not running all the tests?

fyi @flub @divagant-martian @matheus23 this started breaking patchbay tests in the combined branch on degradation level 6 so I will go back to the previous version

With `now + duration`, the PTO fires `duration` after the current poll, regardless of when the last probe was sent. When the PTO cap compresses `duration` below the path RTT (e.g. cap=1875ms vs RTT=1600ms at 800ms one-way latency), consecutive PTO firings overlap with in-flight probes that haven't had time to be acknowledged yet. In patchbay degrade_ladder tests at level 6 (800ms latency, 20% loss, 20% reorder), this causes the client to exhaust its holepunch timeout (15s) with spurious retransmissions before the PATH_CHALLENGE/RESPONSE exchange completes. The server side is less affected because it does not initiate holepunching. Fix: use `max(last_ack_eliciting, now) + duration` so each PTO waits at least `duration` from the last send, giving the peer a full RTT to respond before the next probe.

…meout

This adopts a variant of the algorithm to cap the PTO interval, but keep computing it from the last_ack_eliciting time to keep the computation stable. And adds documentation of why this matters. Also some other small doc and comments improvements around. And fix the SendableFrames::other computation.

flub · 2026-03-26T18:12:58Z

            };

+            let backoff = 2u32.pow(path.pto_count.min(MAX_BACKOFF_EXPONENT));
+            let duration = path.rtt.pto_base() * backoff;


This affects the tail-loss probes during the handshake. There are tests for this and we did not have issues with that. So I have left this computation unchanged.

Adding a comment as to why the handshake is a special case would be nice!

I... I don't know. Other than the existing comment. I mostly left this untouched because it was already untouched. And because it was already always using now + duration which I don't really follow. It possibly is because the space is essentially fixed so you can't get into trouble there.

In theory the handshake has the same issue that when a packet is lost the exponential backoff will grow to large delays, and you probably give up when the idle timeout hits. I think the cap approach should be applicable there as well, you would increase the number of packets sent before giving up. OTOH it is not as needed, this this is mostly about improving how quickly you detect a loss IIUC.

FWIW, I also don't know why this doesn't introduce any jitter in the PTO calculations.

PTAL: I've implemented a version of that also uses the cap during the handshake. This otherwise keeps the same behaviour as was there previously: counting from now. I think that behaviour is fine because when this clause kicks in there only is ever one space that gets used. But it is still weird to me. I also suspect that the previous version was rather sad as well and resulting in way fewer tail-loss probes than would be desired during this time.

flub · 2026-03-26T18:19:08Z

PTAL, I implemented a more gradual cap on the PTO interval. And adjusted the computation so that it computes the correct duration since the last ack-eliciting packet was sent.

i made the changes myself now :)

matheus23

Only nits

matheus23 · 2026-03-30T09:10:16Z

            };

+            let backoff = 2u32.pow(path.pto_count.min(MAX_BACKOFF_EXPONENT));
+            let duration = path.rtt.pto_base() * backoff;


Adding a comment as to why the handshake is a special case would be nice!

Co-authored-by: Philipp Krüger <philipp.krueger1@gmail.com>

n0bot bot added this to iroh Mar 21, 2026

github-project-automation bot moved this to 🚑 Needs Triage in iroh Mar 21, 2026

dignifiedquire force-pushed the fix/pto-cap branch from cac0daa to f6cf1dd Compare March 23, 2026 10:25

dignifiedquire closed this Mar 23, 2026

github-project-automation bot moved this from 🚑 Needs Triage to ✅ Done in iroh Mar 23, 2026

dignifiedquire requested review from flub and matheus23 March 23, 2026 10:27

dignifiedquire reopened this Mar 23, 2026

github-project-automation bot moved this from ✅ Done to 🚑 Needs Triage in iroh Mar 23, 2026

dignifiedquire force-pushed the fix/pto-cap branch from f6cf1dd to a18edde Compare March 23, 2026 10:59

dignifiedquire mentioned this pull request Mar 23, 2026

fix(udp): propagate network-unreachable send errors instead of swallowing them #527

Merged

dignifiedquire added a commit that referenced this pull request Mar 23, 2026

merge: combine #522 and #523 (resolve test conflict)

9dde048

dignifiedquire added this to the noq: iroh v0.98 milestone Mar 23, 2026

dignifiedquire force-pushed the fix/pto-cap branch from 7c42112 to 6d2bcad Compare March 23, 2026 16:16

flub requested changes Mar 23, 2026

View reviewed changes

Comment thread noq-proto/src/connection/mod.rs Outdated

Comment thread noq-proto/src/connection/mod.rs Outdated

Comment thread noq-proto/src/connection/mod.rs Outdated

Comment thread noq-proto/src/tests/multipath.rs Outdated

Comment thread noq-proto/src/tests/proptests.rs Outdated

github-project-automation bot moved this from 🚑 Needs Triage to 🏗 In progress in iroh Mar 23, 2026

dignifiedquire requested a review from flub March 23, 2026 17:50

flub requested changes Mar 24, 2026

View reviewed changes

Comment thread noq-proto/src/connection/mod.rs Outdated

dignifiedquire requested a review from flub March 24, 2026 13:19

divagant-martian reviewed Mar 25, 2026

View reviewed changes

Comment thread noq-proto/src/connection/mod.rs Outdated

Comment thread noq-proto/src/connection/mod.rs Outdated

Comment thread noq-proto/src/connection/mod.rs Outdated

matheus23 previously requested changes Mar 25, 2026

View reviewed changes

flub requested changes Mar 25, 2026

View reviewed changes

Comment thread noq-proto/src/connection/mod.rs Outdated

flub reviewed Mar 25, 2026

View reviewed changes

Comment thread noq-proto/src/connection/mod.rs Outdated

dignifiedquire added 5 commits March 25, 2026 18:58

fix(proto): clarify PTO scheduling comment to match code

29d7a58

CR: apply suggestions

db9b5d3

fixup: doc comments

ccec93b

fix(proto): remove close() workaround now that #538 sets path idle ti…

ce17194

…meout

dignifiedquire force-pushed the fix/pto-cap branch from 185a6e0 to ce17194 Compare March 25, 2026 17:58

flub reviewed Mar 26, 2026

View reviewed changes

docs and format

0c7dee4

flub requested review from divagant-martian and matheus23 March 26, 2026 18:24

flub added 3 commits March 27, 2026 12:57

learn maths

536791c

typos

75629fb

maths!

21893a4

matheus23 approved these changes Mar 30, 2026

View reviewed changes

flub and others added 4 commits March 30, 2026 11:26

use inclusive range for simpler logic

84c9e11

Co-authored-by: Philipp Krüger <philipp.krueger1@gmail.com>

Fix computation of the SLOW_RTT_THRESHOLD

b24e9ef

fix it the right way

24b4c76

cap tail-loss probe interval during handshake as well

d1566b5

matheus23 reviewed Mar 30, 2026

View reviewed changes

Comment thread noq-proto/src/connection/mod.rs

flub added this pull request to the merge queue Mar 30, 2026

github-merge-queue bot removed this pull request from the merge queue due to failed status checks Mar 30, 2026

flub added this pull request to the merge queue Mar 30, 2026

Merged via the queue into main with commit 50faca4 Mar 30, 2026
36 checks passed

flub deleted the fix/pto-cap branch March 30, 2026 16:26

github-project-automation bot moved this from 🏗 In progress to ✅ Done in iroh Mar 30, 2026

divagant-martian mentioned this pull request Apr 8, 2026

wip: client does not open path for nat traversal #572

Draft

Conversation

dignifiedquire commented Mar 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Implementation

References

Uh oh!

github-actions bot commented Mar 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions bot commented Mar 21, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Performance Comparison Report

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Netsim Benchmarks (network simulation)

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Raw Benchmarks (localhost)

Netsim Benchmarks (network simulation)

Summary

Uh oh!

divagant-martian commented Mar 23, 2026

Uh oh!

dignifiedquire commented Mar 23, 2026

Uh oh!

flub left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

dignifiedquire commented Mar 21, 2026 •

edited

Loading

github-actions bot commented Mar 21, 2026 •

edited

Loading

github-actions bot commented Mar 21, 2026 •

edited

Loading

divagant-martian left a comment •

edited

Loading

flub commented Mar 26, 2026 •

edited

Loading