Categorizing Diffs for Review/Transport Call.

Scheffenegger, Richard Richard.Scheffenegger at netapp.com
Tue Apr 21 23:51:26 UTC 2020


I thought I'd categorize the Diffs I have currently pending for review into different bins;

Those which I deem more critical are those with interoperability issues towards other stacks (and RFC violations):
23364
24237
24515
23353

Correctness (RFC compliance)
      ECN:
               23364 (ECN: CWR only with new data) - critical for interoperability with older Linux
               23373 (parallel open case - ECN feature)

      Interoperability:
               24237 (premature window update with outdated SACK - fix alike MacOS and what jtl suggested, by moving upcall to end of input processing; side effect is fewer locking issues) - poor performance with most Linux
               23371 (parallel open case - feature negotiation)
               24515 receive window retraction with window scaling - potential deadlock with older Linux

    Performance
22438 after_idle ssthresh to ¾ cwnd per 2861 (performance)
     Cubic:
               23353 Cubic no minimal cwnd (set to 2 MSS in this diff)
               23655 Cubic base_time start only after slowstart is finished

    Misc:
               21117 optimize TCPCB by replacing unnecessarily large uint64_t with uint8_t and bounded increment
               23160 SACK partial ACK always activate TT_REXMT instead of TT_PERSIST (no longer reproduceable with other cubic fixes) -> abandon?

Feature s:
               23230 ECN+ / ECN++ (new sysctl, default disabled; needed for DCTCP later)
               18624 Improve SACK scoreboard bookkeeping
               18892 SACK PRR
               18985 RFC6675 SACK (rescue retransmission - for transactional IO).

Finally, I have AccECN on my plate, but that partially depends on prior bookkeeping fixes, and finalization of the current draft;

Richard Scheffenegger



More information about the freebsd-transport mailing list