Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
fix(enginenetx): refine the happy-eyeballs algorithm (ooni#1296)
We want to pack attempts in parallel, which we also did before when the interval between attempts was linear. We need to take into account possible congestion, so we should push back exponentially, even though the common case for us is probably censorship (but it is better to do the right thing anyway). So, let's scale exponentially until we reach 30s. After that, it's fine to keep attempts evenly spaces, because 30s is quite definitely a huge interval if we're reasoning in internet time. Also, change the base value used for TLS handshaking to be 900ms rather than 300ms, because a TLS handshake is ~3 round trips. Part of ooni/probe#2531
- Loading branch information