[quicwg/base-drafts] PTO probes are sent too frequently (#3546)

[Tatsuhiro Tsujikawa <[e-mail]>]

It looks like under the current PTO algorithm PTO probes are sent too frequently under certain condition.
To create such a condition, suppose that client never send Handshake packets.
According to the -27 spec, client sends ack-eliciting anti-deadlock packet on pto_count=1.
Then it gets ACK from server for it, which resets pto_count to 0.  After next pto timeout, it sends anti-deadlock packet again.  This process repeats indefinitely until either endpoint gives up handshake.  The problem is that depending RTT, PTO probes are sent too frequently.  With RTT=6ms, it sends PTO probe once per ~33ms.  And both sides do this.  ACK+PADDING packets are also sent without delay in this period if remote endpoint does PTO.   I'm not sure TCP does the similar things, but it sounds like PTO probe is too aggressive.
If both endpoints are implemented properly, this probably does not happen, but in practice, we don't know the remote endpoint is broken in this case, and have no means to check that out.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/3546