[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]
[tor-relays] Does the BEGIN_DIR attack affect user count metrics?
When your relays are experiencing a BEGIN_DIR flood, does it also affect
their user count metrics? For example, in the "average number of
connected clients" graph at https://metrics.torproject.org/rs.html.
User count metrics come from counting directory requests (the
"dirreq-v3-resp" line in metrics descriptors).
https://metrics.torproject.org/reproducible-metrics.html#relay-users
In August and September 2026, there have been unusual, temporary,
apparent increases in users of one of the two Snowflake bridges: just
snowflake-01, not snowflake-02. The country geolocation is "??", which
usually means the Snowflake proxy (or whatever is acting in the role of
a Snowflake proxy and connecting directly to the bridge's WebSocket
port) does not send a client IP address. See discussion and graphs:
https://lists.torproject.org/mailman3/hyperkitty/list/anti-censorship-team@xxxxxxxxxxxxxxxxxxxx/thread/L2XI56NB4GLNG4VFF6VARA64RJBW54OD/
https://opencollective.com/censorship-circumvention/projects/snowflake-daily-operations/updates/2026-august-update
https://opencollective.com/censorship-circumvention/projects/snowflake-daily-operations/updates/2026-september-update
A BEGIN_DIR attack could explain the apparent increases in users,
because it is directory requests that are counted to estimate user
counts. I'm curious if relay operators who have experienced a BEGIN_DIR
attack have also seen apparent increases in user counts.
It's possible that the additional bandwidth engendered by such
an attack went unnoticed at the Snowflake bridge, because the amount of
naturally occurring bandwidth is already so high.
Indeed, I just made a graph using bandwidth-multi.csv from
https://gitlab.torproject.org/dcf/snowflake-graphs/-/tree/3a4213928ed32cef5d7fb49c4b4ccaa1f731d020/
and the "dirreq-write" bytes of the snowflake-01 bridge almost double
during the event of 7 Sep – 20 Sep. (Though there is nothing obvious for
the event of 21 Aug – 23 Aug.) The "dirreq-read", "write", and "read"
bytes do not show any obvious difference. The increase in "dirreq-write"
is from 400 GB to 800 GB / day, whereas the ordinary "write" bytes are
on the order of 30000 GB / day, so the attack bandwidth (if that's what
it was) was negligible in comparison.
_______________________________________________
tor-relays mailing list -- tor-relays@xxxxxxxxxxxxxxxxxxxx
To unsubscribe send an email to tor-relays-leave@xxxxxxxxxxxxxxxxxxxx