08 Sep 2016 Hangout
- Questions arise about where to install tests that come from ports and the answer is likely /usr/local/tests along side /usr/tests.
- The plan to remove flowtable is considered to be a good plan. There remains a bug in IPv6 handling of this.
Jonathan will commit his long changes tonight.
- Why do we have two different ways of sending a RESET?
- Bandwidth limiting in ICMP may not actually work on multi-core. If all the CPUs are busy then the rate limiting may not kick in. rrs@ et al have a patch that should address it.
rstone@ has a patch that correctly merges ECN state on LRO. Discussion revolves around whether or not to flush on ECN rather than propagating the bits to all packets. The correct fix is the Linux GRO model where they do not merge frames, unless the headers are exactly identical. The performance issues get into the problems with mbufs. https://reviews.freebsd.org/D7821
- The TCP_NOOPT option says that all options are dropped from SYN packets after the 3rd try. That is incorrect because MSS is always communicated. We need to make the system consistent and keep MSS always, and also update the documentation.
- rrs@ is planning to commit RACK. jegg@ points out that the code is working (mostly) but that hiren@ is tracking down a locking issue in their version of this code. jegg@ says they are testing with very highly variable RTTs (100ms to 5s) which is a good test for the rack code.