Bufferbloat's "Cerowrt" project

I wanted to point out that our intent is not to fork openwrt - but by our choosing one 100% open source platform for cerowrt, rather than many, we're better able to experiment with new approaches to thorny problems, which can be cycled faster, and solutions then migrate out into the kernel mainline as well as other devices, done by other people.

We've been pushing patches upstream as fast as possible, and they've been pushing patches down to us, at a pretty rapid clip. It's been tons of fun, actually.

As Dave said, CeroWrt is a specific OpenWrt build aimed at the WNDR3700v2, and we are deliberately initially restricting the hardware we target. There are two reasons why we're doing this:

1) we want/need a home router that is fully open, and many/most routers require at least one binary blob someplace. All our experience to date is that there are problems that are not directly bufferbloat, but include bugs and/or naive drivers that delay packets for other reasons. Having a fully open driver stack is necessary to make progress. As Dave pointed out, https://www.bufferbloat.net/issues/216 is a good example of such a situation. Sometimes fixing these problems interact in subtle ways with TCP. Not having full sources to everything would stop progress entirely.
Without a reproducible test platform, attempts to test full solutions for bufferbloat, which need to include AQM, are doomed to spin their wheels.

2) Home routers going forward need to "just work" in ways that they have not in the past, including name services and DNSSEC: you should be able to plug in a named computer into your network and have it "just work", by which I mean a name appears in the global DNS with an IPv6 address. And DNSSEC all the way into the home is necessary: the bad guys are really now out to get us.
Theoretical attacks against name services have turned into real ones in recent months. In part, this is to try to make the router attractive to those who might be able to help with bufferbloat but also because we believe that these are problems we want solutions to in the home routers we run personally.

My experience in taste testing routers over the last year is that OpenWrt comes closest to what I want and need for my home network, and we hope it will be for your home network too. Commercial routers are currently running of order 6 *years* behind the open source community. The other open source router projects are running well behind kernel.org and are therefore also unsuitable for this work: fixes for bufferbloat need to be able to go upstream immediately without lots of effort expended on forward/backward porting.

So to achieve these aims, we've chosen a particular modern abgn router (the WNDR3700v2) as an initial target that is both affordable but one which most LWN readers would find desirable. It has 16 megabytes of flash, which many other home routers do not, so we won't run out of flash space. It's fully open. We'd like to tackle the bufferbloat problem without distractions like drivers with problems we can't fix, and not enough flash for what we're doing.

But also as Dave has said, this initial CeroWrt OpenWrt build isn't quite soup yet, as we chase some of the most egregious bugs uncovered to date. Soon...