feat(iroh): Make poll_writable precise by using NodeMap::addr_for_send

[matheus23]

For now just a PR to look at netsim numbers.

Description

Every send side of a connection gets its own UdpPoller.

If we pass in the Connection::remote_address() to AsyncUdpSocket::create_io_poller in quinn (n0-computer/noq#57) then we can determine which path we're going to use in try_send via NodeMap::addr_for_send within IoPoller::poll_writable.

This allows us to accurately (ignoring race conditions between try_send and poll_writable) report readiness.

Notes & open questions

Needs changes to quinn's AsyncUdpSocket trait.
It's possible to circumvent them, but it's verrrry annoying to do so.
(The trick would be to wrap calls to quinn::Endpoint::connect_with and quinn::Incoming::accept[_with] with a lock and pre-load the remote address somewhere so that MagicSock::create_io_poller can fetch it back when it's called through connect_with/accept[_with]. I couldn't get that to work easily without panicing 😬 )

Change checklist

• Self-review.
• Tests if relevant.

[github-actions]

Documentation for this PR has been generated and is available at: https://n0-computer.github.io/iroh/pr/3266/docs/iroh/

Last updated: 2025-05-20T15:12:46Z

[github-actions]

Netsim report & logs for this PR have been generated and is available at: LOGS
This report will remain available for 3 days.

Last updated for commit: 1150c5d

[matheus23]

This PR:

test	case	throughput_gbps	throughput_transfer
iroh	1_to_1	1.38	1.38
iroh	1_to_3	3.24	3.25
iroh	1_to_5	4.98	4.98
iroh	1_to_10	6.52	6.53
iroh	2_to_2	2.71	2.72
iroh	2_to_4	3.93	3.93
iroh	2_to_6	5.68	5.69
iroh	2_to_10	7.92	7.93

Latest PR to main:

test	case	throughput_gbps	throughput_transfer
iroh	1_to_1	1.37	1.37
iroh	1_to_3	4.06	4.07
iroh	1_to_5	6.93	6.94
iroh	1_to_10	12.14	12.16
iroh	2_to_2	3.06	3.07
iroh	2_to_4	5.83	5.84
iroh	2_to_6	9.26	9.27
iroh	2_to_10	13.74	13.76

[matheus23]

I guess it clearly makes a difference? 😅

[matheus23]

Don't think I'll merge this near-term.
Needs more though/figuring out why performance is so much worse.
And I don't want to pollute the pull requests tab, so closing.

[matheus23]

This PR after rebasing on main:

test	case	throughput_gbps	throughput_transfer
iroh	1_to_1	1.41	2.79
iroh	1_to_3	4.50	9.50
iroh	1_to_5	7.15	14.34
iroh	1_to_10	11.85	20.35
iroh	2_to_2	2.78	5.48
iroh	2_to_4	5.49	10.68
iroh	2_to_6	9.01	19.23
iroh	2_to_10	13.77	26.81

main:

test	case	throughput_gbps	throughput_transfer
iroh	1_to_1	1.45	2.95
iroh	1_to_3	4.58	9.87
iroh	1_to_5	6.59	12.30
iroh	1_to_10	12.27	21.67
iroh	2_to_2	2.59	4.78
iroh	2_to_4	5.89	12.19
iroh	2_to_6	8.43	16.79
iroh	2_to_10	14.02	27.93

[matheus23]

Closing this in favor of #3279 (and it's using n0-computer/noq#67 which enables the same thing).