Join Nostr
2026-10-08 03:03:47 UTC

MoneroResearchLab on Nostr: rucknium re-ran the unreachable-node simulations for 50 in-sim hours because the ...


rucknium re-ran the unreachable-node simulations for 50 in-sim hours because the 26-hour runs never let white_lists reach steady state. The 26-hour plot does not approach the 1,000-peer threshold; the 50-hour run almost does, as 1,000 adversary peers are slowly absorbed into the unreachable node’s white_list. The qualitative result is unchanged. gingeropolous helped speed the sims: 15 hours of wall-clock for 50 hours in-sim. Reproducible analysis, then a PR with the suggested changes, is expected within about 24 hours. Full reproduction needs a ~1TB-RAM machine; raw sim data will be included, and a scaled-down sim may follow. Status-quo runs sometimes hold a sustained 12/12 eclipse and sometimes only an intermittent one, which rucknium attributes to sim randomness and treats as a reason to run multiple trials. Prior confidence intervals: https://gist.github.com/Rucknium/ba230a2bab8a446d0997b7b002bf3af6.

rucknium: 5. Shi et al. (2026) "Are Unreachable Nodes Truly Safe? Fully Eclipsing Monero’s P2P Network!" (https://arxiv.org/pdf/2609.10260)
rucknium: I re-ran simulations for a longer 50 hours in-sim time. I wasn't satisfied with the 26-hr simulations because the white_lists didn't reach a steady-state equilibrium.
rucknium: (white_list_unreachable.png)
rucknium: ^ This is the 26-hour run. You can see that the red line on both plots does not go up near the 1,000-peer threshold.
rucknium: (white-list-unreachable.png)
rucknium: The 50-hr run almost reaches that. There are 1,000 adversary-controlled peers. It takes a while for them to be absorbed by the unreachable node's white_list.
rucknium: But anyway, the same result happens. Just wanted to make sure it worked with the longer sim time.
rucknium: (outbound-adversaries-unreachable.png)
rucknium: gingeropolous helped make the sims run faster. We are at 15 hours of wall-clock time for this 50-hr in-sim time simulation.
rucknium: I've been working on making the analysis reproducible. It is sometimes surprising how much time it can take to gather scattered scripts and data and write understandable reproduction instructions. But if it's not reproducible, it's not science! Once I'm done with that, which should be within 24 hours, I'll make the suggested changes as a PR.
rucknium: Well, it won't be "easily" reproducible from scratch if you don't have a 1TB-RAM machine. But I will include the raw sim data. I may try to create a scaled-down sim that should show the results, after making the PR.
rucknium: That concludes my update on this agenda item.
rucknium: Actually, one more thing. gingeropolous and I notice that sometimes the status quo sim results in a sustained 12/12 eclipse, and sometimes only intermittent 12/12 eclipse. I think that's because of randomness in the sim. That just brings up why running the sim multiple times can be important. In this simulation case, the outcomes for the status quo and proposed defense simulations are so different that I think doing statistical inference with multiple simulations each probably isn't completely necessary.
rucknium: This is where I did multiple simulations and estimated confidence intervals on outcomes, previously: https://gist.github.com/Rucknium/ba230a2bab8a446d0997b7b002bf3af6
jeffro256: Ruck is CGO: chief graph officer

https://libera.monerologs.net/monero-research-lab/20261007#c713326
Shi et al. (2026) "Are Unreachable Nodes Truly Safe? Fully Eclipsing Monero’s P2P Network!": https://arxiv.org/pdf/2609.10260
nevent1qqswyt808xtymr98k5ghuqp9l7lvsgmp7jn9euxq7cfjecnrfq4p9nc2955qd