Hey Hiro, I have now written a project [1] to collect the relay count from [2]. Following observations: 1. Most spikes at [3] (as you suggested) appear to be measurement errors 2. There are 4 spikes visible at [4]: a.) Positive spike in 12/2014, for around 11 hours b.) Negative spike in 12/2021, only one hour c.) Negative spike in 01/2026, only one hour d.) Positive spike in 01/2026, around 11 hours I will have to check if the respective consensus files for (b) and (c) show some signs of corruption (or failed consensus). The relevant logic for parsing the count for running relays can be found at [8]. 3. ) The relay count shows some 24-hour (12h and 30d as well) periodicity [5]. I'm guessing that this is could be due to individuals who operate relay servers on their personal machines. Or maybe certain relay servers crash on high traffic?
Relay operators are responsible for updating their system yes. What is your concern in this case? My concern is a "trust bottleneck" in the release process. Tor appears to only be signed by a maximum of 3 developers [6] and signatures published on [7]. I imagine that the risk of supply chain attacks could be reduced, if there would be a way to verify a release over the directory authorities. That would require the builds to be reproducible and the DA operators to have the capacity to review the source code.
Best, ttlns [1] https://codeberg.org/ttlns/torcollector_relays_stats_test [2] https://collector.torproject.org/archive/relay-descriptors/consensuses/ [3] https://metrics.torproject.org/networksize.html?start=2010-01-01 [4] https://codeberg.org/ttlns/torcollector_relays_stats_test/src/branch/main/as... [5] https://codeberg.org/ttlns/torcollector_relays_stats_test/src/branch/main/as... [6] https://support.torproject.org/little-t-tor/getting-started/verifying/#bsd-l... [7] https://www.torproject.org/download/tor/ [8] https://codeberg.org/ttlns/torcollector_relays_stats_test/src/branch/main/sr...