TCP Hangout 15 Nov 2015
Hiren
- Committed RFC 6675 support
- Working with Randall on SCK + loss issues
- Per SACK hole DUPACK counter
Randall
- Needs the indirection code reviewed
Will test on 10.X at NetFlix
- The overhead of indirection might not matter on other architectures.
- RackTCP (related to QUICC) can be done as a KLD.
gnn
- Looking into how we handle out of sequence RST based on information from Richard Clayton at Cambridge University
The RackTCP stuff requires SACK for a scoreboard.
97% of the web traffic seen at NF is SACK based.
Hiren has been reviewing CUBIC
Matt Macy has untested changes for driver back pressure.
MPTCP still needs review (email ongoing with Nigel)
Peter Sewell and Hannes may look at MPTCP.
George to reach out to Nigel about spending some time at Cambridge.
Thanksgiving week meeting canceled. Resume first week of December.