We have conducted a few more experiments to check some parameters of the
attack. First, we tested the stability of the fingerprint (eight first-hop
connections). We were running our bitcoin client for about 10 hours. During this time 3
nodes out of eight have disconnected. Analyzing the data from
http://getaddr.bitnodes.io/ we see that after two hours at most one node
would disconnect in average. So for the typical duration of a client session the
fingerprint is very stable.
The second point we wanted to check was regarding the usage of VPN. Indeed
it will protect user's IP, but we wondered what would happen with the
stability of a fingerprint. I've made a couple of experiments on the
testnet, connecting to it:
1) via public free VPNs (vpngate.net);
2) via a non-free one (AirVPN).
3) via my own VPN server.
For the cases 2) and 3) the stability of fingerprint was the same as if no
VPN was used.
For the case 1) connections to entry nodes were dropped
from time to time (about every 20 mins for the main net and about every
few minutes for the testnet due too absence of traffic) by the VPN servers.
It's likely that free VPN servers were set with small inactivity timeouts
and some limits for connection durations.
We are currently working on a countermeasure. In particular we wonder how
reducing from eight to four entry nodes would effect the bitcoin network.
We also think to decay the connections based on the number of transactions
(for example renewing half of the entry node connections after each
transaction). It might be interesting to include a proof of work for
creation of connections, but this would create a compatibility issue.
