Set "AuthDirVoteStableGuaranteeMinUptime 0".

[sporksmith]

This should ensure that all relays get the Stable flag, fixing #123.

[sporksmith]

At first glance of the CI output, this change appears to substantially improve simulated network performance. This is probably a good thing, but I wonder whether it calls into doubt previous simulation results and/or means some other parameters were compensating for this difference and now need to be readjusted.

https://github.com/shadow/tornettools/actions/runs/28383130535/artifacts/7957053330

e.g.:

circuit_build_time.exit.pdf

client_goodput.exit.pdf

client_goodput_5MiB.exit.pdf

It's been a while though since the "golden" dataset has been updated. I'll do a CI run on main to see whether whether these differences are already present without this PR.

[sporksmith]

A fresh no-op run on main has similar differences from the "stable" baseline. https://github.com/shadow/tornettools/actions/runs/28385211794/artifacts/7957890090

I'll update the baseline, and then rebase and rerun this one.

[robgjansen]

Hmm!! It does seem like we're going to want to adjust some other parameters to do a better job of matching public Tor. That is probably going to mean adding more circuits and load (concurrent downloads), and maybe adjusting the defaults once we find a good sweet spot.

I'd suggest we also figure out #122 and apply that change too before minting a new "golden" baseline.

Note: I was/am planning to put in more work on this. I already have some tools for containerizing the benchmark tools (using apptainer) so it will be easier to migrate it to a different runner, and it will be much easier to mint a new "golden" baseline checkpoint whenever we want without having to go through our currently complicated cache update process. Tweaking tornettools params and iterating is always part of creating a new baseline, but I think it will be even more important after this change and #122.

Do you want this change urgently or should I continue down my slower but maybe more measured path?

[robgjansen]

Oops, I missed your most recent comment. I guess the jury is still out about the effects of this one. But I think my comment still applies generally.

[sporksmith]

I've already gone ahead with putting together a baseline update: #125

Assuming you have no objection to merging that, I'll go ahead and re-run this one after rebasing to satisfy my own curiosity if nothing else :)

No rush for merging the actual torrc change though - I'm happy to let you handle a more thorough evaluation and tuning.

[sporksmith]

With the baseline-data update the differences here don't look as drastic. They could just be the "normal" lack of determinism we've seen here. https://github.com/shadow/tornettools/actions/runs/28390499336/artifacts/7959799172

I'll go ahead and close this for now, pending a more thorough investigation