03 December 2015 Meeting Minutes
Jonathan is not waiting on anyone for further comments on his commits and will commit his work.
Randall's code is ready to go in but is awaiting testing. It may affect other architectures more negatively than amd64. AdrianChadd ought to run this on top of a MIPS like architecture. Test on BBB as well.
Patrick has put up a review of server side fast open: https://reviews.freebsd.org/D4350
gnn has talked to Mellanox about their packet pacing code which has led to further discussions on pacing and back pressure. Matt is not convinced that the timers will work well in Mellanox. Chelsio's solution seems to be more mature on the hardware side. Matt has a review up at https://reviews.freebsd.org/D4295. Also to be reviewed https://reviews.freebsd.org/D4294, https://reviews.freebsd.org/D4293, https://reviews.freebsd.org/D4292 as they relate to TCP constants and timing.
gnn will talk to Nigel about getting a review of MPTCP into Phabricator.
Hannes and Peter do point out that their model would have found the RST out of window issue discussed previously.
gnn will ask Hannes whether they are using the DTrace probes to run their model rather than the old TCP_DEBUG #define
Mike Karels has a review to re-add the route cache for TCP https://reviews.freebsd.org/D4306 Matt has a patch that is not in review but he will push that up and relate it to Mike's patch. Bjoern also has a copy of the projects branch patch. gnn will explain the different between routers and hosts to those who complain about this change.
Bjoern suggests breaking the patch down so that we take it in chunks. Ultimately we may need to re-add the code that passes routes through the stack which were removed previously.
Matt has a set of recommendations in email relating to incast on the transport@ mailing list. Others should comment.