Circuit creation timeout
karsten.loesing at gmx.net
Thu Jul 5 17:00:59 UTC 2007
-----BEGIN PGP SIGNED MESSAGE-----
Johannes Renner and me have conducted some statistics on circuit setup
(well, he collected the data and I analyzed them). We chose default
settings for Tor but disabled using entry/guard nodes.
Results show that the default timeout of 60 seconds is far beyond
reasonable circuit establishment times. Median for a 3-hop circuit setup
is 4.36 secs with IQR of 6.01 secs, resp. average of 7.78 secs with std.
dev. of 9.20 secs. 27% of all circuit creation attempts fail, of which
5% (9%, 13%) fail at creating the first (second, third) hop. The medians
for the three hops are 0.58 secs for the first, 1.23 secs for the
second, and 2.23 secs for the third hop; the IQRs are 0.66 secs for the
first, 1.80 secs for the second, and 3.01 secs for the third hop. If
someone is interested in the original data or the SPSS output, I could
send it by mail or upload it somewhere.
A first discussion on #tor resulted in the following arguments:
- - One could reduce the default timeout for circuit creation to 25 secs.
- - In case of a crappy network, one could back off after a certain number
of failures to 60 secs. Therefore, some kind of history and dynamic
timeout determination of the timeout would be required.
- - One could keep a history of previous connection times to better cope
with the network connection -- if that should make a big difference
which we did not measure, yet.
- - One could have distinct timeouts for each hop, because there is no big
use in continuing circuit creation after the first hop requires a large
share of the overall timeout. Overall creation might not succeed and the
circuit might be slow.
- - Are there anonymity issues in cutting down the circuit creation time?
IMHO not, if some value bigger than the 0.9 or 0.95 quantile is chosen.
Any ideas on this?
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org
-----END PGP SIGNATURE-----
More information about the tor-dev