TCP Cubic: Recalculate cubic cwnd for every ack.
ClosedPublic
Actions

Authored by rscheff on Aug 13 2020, 8:17 PM.

Details

Reviewers

cc
tuexen
rrs
rgrimes
kbowling

Group Reviewers

transport

Commits

rS365295: MFC r364354: TCP Cubic: recalculate cwnd for every ACK.
rS364354: TCP Cubic: recalculate cwnd for every ACK.

Summary

Since cubic calculates cwnd based on absolute
time, retaining RFC3465 (ABC) once-per-window updates
can lead to dramatic changes of cwnd in the convex
region. Updating cwnd for each incoming ack minimizes
this delta, preventing unintentional line-rate bursts.

Diff Detail

Repository

rS FreeBSD src repository - subversion

Lint

Lint Not Applicable

Unit

Tests Not Applicable

Event Timeline

rscheff created this revision.Aug 13 2020, 8:17 PM

Herald added a reviewer: transport. · View Herald TranscriptAug 13 2020, 8:17 PM

Herald added a subscriber: melifaro. · View Herald Transcript

rscheff requested review of this revision.Aug 13 2020, 8:17 PM

Harbormaster completed remote builds in B32979: Diff 75800.Aug 13 2020, 8:17 PM

tuexen accepted this revision.Aug 13 2020, 8:21 PM

This revision is now accepted and ready to land.Aug 13 2020, 8:21 PM

cc requested changes to this revision.Aug 14 2020, 1:49 PM

cc added inline comments.

sys/netinet/cc/cc_cubic.c
135–137 ↗	(On Diff #75800)	Shall have the comment updated accordingly.

This revision now requires changes to proceed.Aug 14 2020, 1:49 PM

Verified this patch here:
https://github.com/ccui1000/freebsd-tcp/wiki/test-D26060

Looks good to me.

always use absolute cubic_cwnd value
update cubic cwnd comment to reflect new logic

Harbormaster completed remote builds in B33049: Diff 75958.Aug 18 2020, 6:00 PM

cc added inline comments.Aug 18 2020, 6:51 PM

sys/netinet/cc/cc_cubic.c
135–137 ↗	(On Diff #75958)	The SUMMARY part of this change already says this. Can we remove this sentence "During the convex region of the cubic curve, cwnd steps can otherwise become very large and cause dramatic line-rate bursts." ?