01 December 2016 Hangout Notes
- hps@ could not make the meeting but has handed off the pacing conversation to gallatin@. The remaining blocking issue for using this was the problem with not returning any indication of a transmit failure. hps@ will update the code to return a unique code for that case and the code review will move forwards. Another sticking issue is having multiple types of NICs in a single LAGG, but that is not a blocking issue.
- gnn@ and rwatson@ met with Neal Cardwell about packetdrill and Neal has committed to upstreaming all of Google's TCP tests and dual licensing them.
The issue previously seen with RTO should be addressed by this patch: https://reviews.freebsd.org/D8556
- rwatson@ asks about using libuinet as a way of testing TCP in a more deterministic fashion.
tuexen@ would like this change to FIN handling reviewed: https://reviews.freebsd.org/D8667
jegg@, hiren@ and gnn@ have read this paper on high RTTs http://dl.acm.org/citation.cfm?id=2815704 which indicated that RTTs are far worse than one might expect and are increasing over time. This is important because it means that many assumptions of bugs in the stack might be incorrect. Separating RTT bugs from network issues remains a key part of debugging these issues.