Transport Meeting 20 Apr 2017
We discussed https://reviews.freebsd.org/D10272 as a group and decided that this patch should be accepted.
jtl@ opened two reviews for ICMP rate limit code and would like review https://reviews.freebsd.org/D10387 https://reviews.freebsd.org/D10388
- jtl@ has new code to make the netstat code more efficient and will put up reviews soon
- gnn@ is going to push on the packetdrill folks to start accepting our changes
- rrs@ is ready to start putting Rack and BBR code into HEAD. The code depends on jtl@'s black box code.
- gnn@ would like to take a look at the black box code before accepting it into the tree. rrs@ to follow up with gnn@ on traces and the new UI they've written.
- rrs@ has structured the inpcb/tcpcb to minimize cache hits on the structures for improved performance. gnn@ suggests getting those diffs to bz@. the initial performance improvement numbers are quite promising as they are now serving 97Gbps and maintain 50% free CPU.
- gnn@ asks about BBR with other workloads. rrs@ says that they'll be working further to address any sharp edges.
- gnn@ will make sure there is a Transport session at BSDCan
- hiren@ is experimenting with avoiding pacing with small files on BBR.
glebius@ is continuing his work on improving the performance of listening sockets. The only sub-system where this might be an issue is SCTP but tuexen@ believes that there is no blocking problem. https://reviews.freebsd.org/D9770
hiren@ asks if the Rack code has jtl@'s long fixes. jtl@ will check but does point out that they have not seen any issues in the system at NetFlix.