03 November 2016 Hangout Notes
- gnn@ and tuexen@ have agreed to move the tcp-testsuite from tuexen@'s personal account to the freebsd-net org
- hps@ brings up the modification of the mbuf for use in packet pacing. There are three choices: use the rcvif in the tx direction, Drew suggests a union and a macro to hide it for backwards compatibility. julian@ has disagreed but as he is not involved heavily in the NIC drivers we have agreed to allow hps@ and np@ along with other vendors to make the decision. The rcvif will wind up being used as a hardware context for the TX direction.
- hiren@ has committed the patch that prevents the stack from going back to 1 MSS on loss. When loss happens we try to follow the congestion control (CC) module's recommendation. People tracking HEAD should look for issues resulting from this change. hiren@ plans to MFC this in one month.
- hiren@ wonders about the alignment of cwnd to the mss boundary. jtl@ says that rrs@ has seen papers that talk about this in segments or bytes. Linux is very packet/segment based and so the issue does not come up. rrs@ has said that this may not be necessary in our stack.
rstone@ talks about how to properly spread connections over CPUs and that the current current RSS code does not work as well as we'd like. There is a review of some old code that might be useful to look at (https://reviews.freebsd.org/D806) but what we also need is a higher level API that addresses the issue.
teuxen@ asks for a review of https://reviews.freebsd.org/D8371