Under what conditions if any would Bitcoin v2 P2P traffic be indistinguishable from other encrypted TCP traffic
BIP-342 supports traffic shaping, which means implementations are able to make the traffic patterns (i.e. timing and sizes of packets) resemble whatever they like, at the cost of needing to add dummy bytes, delay traffic when there are spikes, and possibly drop traffic when it persistently exceeds the traffic patterns that are being mimicked.
However, this is an implementation decision, and as far as I know, there are none that implement it. There is no such thing as “random traffic”. All traffic has patterns. The question is just what patterns to mimic (or not).
What remaining properties of the v2 transport such as packet sizes handshake structure or message timing still allow traffic classification
That is what traffic patterns are. All actually sent bytes are uniformly random (which itself is distinguishable from most traffic, but a neutral common denominator to target).
BIP-324 allows hiding almost all patterns, except for a few very weak ones in the initial handshake. The initiator always sends between 64 and 4159 bytes at once (which may consistute one or more TCP/IP packets). The receiver waits until at least one byte is received, and then also sends between 64 and 4159 bytes at once. Footnote 19 explains how these conditions can be further relaxed in the future in a backwards-compatible change, if need be.
But again, the actual extent to which traffic shaping takes place is implementation dependent.
- Does BIP 324 meaningfully approach traffic obfuscation or does it mainly defeat simple protocol signatures while remaining classifiable through traffic analysis
In practice, today’s implementations only defeat protocol signatures that involve patterns in the sent bytes, not in traffic.
