3.3.8-10 and -11 were not as good as I'd hoped. I have been doing continuous integration while on this trip but have not been in a position to test anything. And after -11 - well, I'm not putting out anything without more serious testing.
The biggest issue we have is that there are multiple attempts in motion to improve codel's behavior under load. We have a half dozen competing ideas that all need simulation and testing. Nearly everybody involved in the codel effort is at this upcoming ietf in vancouver, so it is my hope that we can get together and get a set of common simulations against the algorithmic variants, and decide on something then. So my plan is to get a cerowrt build out sometime next week. (aug 1-3) An issue I fixed in the cerowrt source tree today is ipv6 default routes weren't making it out via babel - It's apparently fixed now, (thx to a patch from denis and now in the net-next linux tree). Pretty much everything else is now stable - for example, dnsmasq 2.62 with AAAA reverse DNS and dhcpv6 support is now in openwrt mainline. As things grow stabler in openwrt/cerowrt, the linux mainline tree continues to advance. I'd like to backport Eric Dumazet's TCP small queues work to this release of cerowrt, if I get time. This looks very promising to reduce host bufferbloat still further: http://lwn.net/Articles/506237/ - and de-heisenbug some tests. BUT THAT'S IT. For example: I am not planning on adding TCP fast open support on this go-round. I do hope to ensure that TCP fast open can work, and perhaps put means in play to measure it working or not, but that's it. Anyway, I got off all my mailing lists while I took this break. So I am out of touch. And refreshed. If you have any bug fixes for cerowrt let me know soonest. -- Dave Täht _______________________________________________ Cerowrt-devel mailing list [email protected] https://lists.bufferbloat.net/listinfo/cerowrt-devel
