Think of more research questions that we can answer with statistics
To decide what HS statistics projects to do for SponsorR, we've been following the approach:think of all statistics we could collect and decide whether we should or not do them
. This approach has lead to various confusions on whether a statistic is useful to us or not.
To make it easier to reason about the usefulness of statistics, we decided to take a different approach which is Enumerate useful research questions we want to answer, and collect statistics that could help us answer them
.
Such research questions could be:
- Is 6 HSDirs the best number of HSDirs?
- How long should HSDirs cache descriptors?
- Are IPs stable or are they subject to churn? How often do HSes change IPs?
- Do we want more preemptive circuits on hidden services?
- Do hidden services fail easily because their sole guard fails?
This ticket is about discussing ideas and figuring out what statistics we need to gather in the end. More ephemeral work happens at: https://etherpad.mozilla.org/psNv5z99Z5