Why I am doing this
Repeating myself endlessly about bufferbloat is more boring for me than you, believe me!

In addition to cerowrt…

I’m also the co-founder of the bufferbloat project, the implementer of codel, contributor to a whole bunch of internet drafts and rfcs, and the and the designer or architect of a whole bunch of the predecessors to other embedded software you use every day… I worked on early linux based handhelds, and during my time at montavista software especially I worked on a metric ton of things that are now everywhere, examples: if you use a panasonic tv, or see an x-windows prompt on an airline seatback, those are my fault. Using sip or asterisk for voip? partially my fault. I worked really hard on stablizing the toolchains for mips, ppc, and arm platforms during that era, also, which is what has made the later explosion of linux into all the embedded spaces possible.

I am NOT taking sole credit! Lordy there were 10s of thousands of people involved in taking linux from 0 embedded devices in 1996 to billions in 2015. I know and worked with dozens and dozens of people that really deserve more credit than they get, for the world as it exists today.

My main claim to fame before bufferbloat was that the creation of the entire home gateway and wifi router industry is partially due to my efforts in the 90s and early 00s. [1]. Not that anyone in the industry bothers to send a christmas card! I do sometimes wish I had the same kind of pull I had in the early 00s. CTOs and VCs used to call me back.

But I can point to things like the above and below that sprang directly from my efforts.

These days I tend to think of the current sad state of home gateways and wifi as more my fault [2]. I should have tried to standardize wondershaper [3], I should have fought harder against wifi packet aggregation when it was first proposed in 04, and I for damn sure should have tried to get it right in 06. Instead I was happily retired and living on a beach in south america while the edge of the internet went to hell [4]. I was annoyed at 802.11e EDCA, but I didn´t even notice how screwed up 802.11n was under contention until late 2010. Nor did I realize that home gateways were going to hell, particularly after the great recession closed all the embedded linux shops like montavista and embedded alley and so on that were doing the good bringup work.

So me and my merry band of some original internet pioneers, and a ton of volunteers, and a few clued companies like google, redhat and comcast, have been trying to fix all that, ever since 2010. [5]. I got sucked into doing the cerowrt portion of the bufferbloat project because it was obvious we needed to be implementing tons of new code on deployable hardware, and nobody else stepped up to the plate that had sufficient spare time to drive it to completion.

My reaction when Vint Cerf, Jim Gettys, Dave Reed, Paul Vixie, Stuart Cheshire, Van Jacobson, or esr ask me to do anything, and offer me a spot on their couch(es) to do it - my response is… “yes, I’ll get on it - when do you need it by?”. [6]

What would your reaction be if even one of these people asked a favor of you? Why don’t the internet megacorps of the world accord these people the same respect - especially when they lay out exactly what they have been doing wrong, and exactly, how fix it, and fall all over themselves to make the technologies available for free? Jeebus.

I would rather dearly like to go back to my beach, but there is one major thing left to do in addition to crossing the chasm, fixing wifi. I am fully aware it is going to take another 4-5 years to flush the bad products out of the marketplace.

Anyway… rant over….

The only advantage to cerowrt 3.10.50-1 remaining is that we have been breaking records for stability and uptime. [8]. I think those sort of records are also being achieved across the board in the openwrt mainline so it is not very important. Otherwise every important feature and technology developed in cerowrt and of interest to normal users is in openwrt chaos calmer, and the linux mainline kernel, and downstream in products that are tracking that stuff closely, unlike :ahem: dlink.

These included vastly better ipv6 support, DNSSEC support for dnsmasq, better wifi drivers, and the newly developed fixes for bufferbloat (debugged HTB, BQL & fq_codel - and as tests all the other proposed aqm variants such as pie are in it). I am really, really happy, in particular to have made Vint happier with ipv6, paul with dnssec, and jim with fq_codel. The preliminary ietf homenet support is not really there, though. Sigh. [9]

So there is now nearly no point in using cerowrt. Find a currently shipped ath9k based product on openwrt´s supported list, put on barrier breaker or chaos calmer and go to it. And/Or nag d-link to catch up. :) I do not as yet have an opinion on the DGL-5500’s overall quality, or its chipset, or its firmware sorry. Maybe its great now. I am easy! But given the unbelievable number of improvements to linux since 2.6.36 I have grave doubts. I will go test it this week.

The previous target of cerowrt´s efforts had, among other things, been flown to 120,000 feet [7] which made me feel comfortable about using it for my userbase. I plan to fly whatever hardware we choose for the next iteration of cerowrt to the same height, or higher - and bake it, freeze it, etc - before deciding to use it.

I am delighted (as cerowrt the research project became a success disaster… ) that I can now point people that want to develop cool new products to multiple more polished, stabler, upstream distros (like openwrt, dd-wrt, ipfire, gargoyle, etc), and leave us researchers alone to go back to finding solutions (and breaking stuff!) for the remaining problems we have identified.

Whenever cerowrt resumes development (april maybe, funding willing) the hope is to implement all the solutions we have come up for wifi’s problems on some more modern piece of hardware. IF the DGL-5500 running openwrt is suitable, and stable enough, awesome. I mentioned another dlink product we are considering in an earlier post, it really depends on how deep we can get our hands into the firmware. It is going to ath9k based for sure for one side of the project and we are still in the final selection process for the other chipset.

I had a bit of fun writing this rant, and getting some of this out of my system, and of the links below, I really, really wish, more people had a deeper intuitive understanding of how wifi really works (second half of my talk at mit) [2], because the world is desparately short on people that can fix it. Wifi is really cool, I am looking forward to making it better, and hope more people can contribute to the process of making the edge of the internet better, for everyone.

[1] http://the-edge.blogspot.com/2010/10/who-invented-embedded-linux-based.html [2] https://www.youtube.com/watch?v=Wksh2DPHCDI&feature=youtu.be “see second half of MIT talk on what´s wrong with wifi” [3] http://www.bufferbloat.net/projects/cerowrt/wiki/Wondershaper_Must_Die [4] https://gettys.wordpress.com/2013/07/10/low-latency-requires-smart-queuing-traditional-aqm-is-not-enough/ [5] http://cero2.bufferbloat.net/cerowrt/credits.html [6] http://esr.ibiblio.org/?p=4196 “holding up the sky” [7] http://snapon.lab.bufferbloat.net/~d/spacerouter.JPG [8] https://lists.bufferbloat.net/pipermail/cerowrt-devel/2015-February/004022.html [9] https://datatracker.ietf.org/wg/homenet/documents/ - homenet is attempting to standardize fixes for mdns and ipv6 enabled home gateways

February 19, 2015
1257 words