[Cerowrt-devel] BBR congestion control algorithm for TCP innet-next
alan.christopher.jenkins at gmail.com
Wed Sep 21 03:32:12 EDT 2016
On 20/09/16 21:27, dpreed at reed.com wrote:
> I constantly see the claim that >50% of transmitted data on the Internet are streaming TV. However, the source seems to be as hard to nail down
I don't think the source is hard to identify. It's Sandvine press
releases. That's what the periodic stories on Ars Technica are always
> as the original claim that >50% of Internet traffic was pirated music being sent over bittorrent.
> You recently repeated that statistic as if it were a verified fact.
> I remember that in the early days of WiFi DSSS availability the claim was repeatedly made from podiums at conferences I attended that "the amount of WiFi in parking lots on Sand Hill Road [then the location of most major Silicon Valley VC firms] had made it so that people could not open their car doors with their remote keys". This was not intended as hyperbole or a joke - I got into the habit of asking the speakers how they knew this, and they told me that their VC friends had all had it happen to them...
> Propaganda consists of clever stories that "sound plausible" and which are spread by people because they seem to support something they *wish* were true for some reason.
> I suspect that this 70% number is more propaganda of this sort.
> In case it is not obvious, the beneficiaries of this particular propaganda are those who want to claim various things - that the Internet is now just TV broadcasting and thus should be treated that way (Internet Access Providers should select "channels", charge for allowing them through to customers, improving the "quality of programming" and censoring anything offensive, as just one example)
> So I am extremely curious as to an actual source of such a number, how it was measured, and how its validity can be tested reproducibly.
> Some may remember that the original discovery of "bufferbloat" was due to the fact that Comcast deployed Sandvine gear in its network to send RST packets for any connections that involved multiple concurrent TCP uploads (using DPI technology to guess what TCP connections to RST and the right header data to put on the RST packets).
> Their argument for why they *had* to do that was that they "had data" that said that their network was being overwhelmed by bittorrent pirates.
> In fact, the problem was bufferbloat - DOCSIS 2.0 gear that was designed to fail miserably under any intense upload. The part about bittorrent piracy was based on claimed measurements that apparently were never in fact performed about the type of packets that were causing the problem.
> Hence: I know it is a quixotic thing on my part, but the scientist in me wants to see the raw data and see the methods used to obtain it.
> I have friends who actually measure Internet traffic (kc claffy, for example), and they do a darn good job. The difficulty in getting data that could provide the 70% statistic is *so high* that it seems highly likely that no such measurement has ever been done, in fact.
> But if someone has done such a measurement (directly or indirectly), defining their terms and methodology sufficiently so that it is a reproducible result, it would probably merit an award for technical excellence.
> Otherwise, please, please, please don't lend your name to promulgating nonsense, even if it seems useful to argue your case. Verify your sources.
> On Monday, September 19, 2016 4:26pm, "Dave Taht" <dave.taht at gmail.com> said:
>> ok, I got BBR built with net-next + v2 of the BBR patch. If anyone
>> wants .deb files for ubuntu, I can put them up somewhere. Some quick
>> I haven't got around to testing cubic vs bbr in a drop tail
>> environment, my take on matters is with fq (fq_codel) in place, bbr
>> will work beautifully against cubic, and I just wanted to enjoy the
>> good bits for a while before tearing apart the bad... and staying on
>> fixing wifi.
>> I had to go and rip out all the wifi patches to get here... as some
>> code landed to the ath10k that looks to break everything there, so
>> need to test that as a baseline first - and I wanted to see if
>> sch_fq+bbr did anything to make the existing ath9k driver work any
>> On Sat, Sep 17, 2016 at 2:33 PM, Dave Taht <dave.taht at gmail.com> wrote:
>>> On Sat, Sep 17, 2016 at 2:11 PM, <dpreed at reed.com> wrote:
>>>> The assumption that each flow on a path has a minimum, stable RTT fails in
>>>> wireless and multi path networks.
>>> Yep. But we're getting somewhere serious on having stabler RTTs for
>>> wifi, and achieving airtime fairness.
>>>> However, it's worth remembering two things: buffering above a certain level is
>>>> never an improvement,
>>> which BBR recognizes by breaking things up into separate bandwidth and
>>> RTT analysis phases.
>>>> and flows through any shared router come and go quite frequently on the real
>>> Very much why I remain an advocate of fq on the routers is that your
>>> congestion algorithm for your particular flow gets more independent of
>>> the other flows, and ~0 latency and jitter for sparse flows is
>>>> Thus RTT on a single flow is not a reasonable measure of congestion. ECN marking
>>>> is far better and packet drops are required for bounding time to recover after
>>>> congestion failure.
>>> Aww, give the code a chance, it's only been public for a day! I
>>> haven't even got it to compile yet!
>>> I think it is a *vast* improvement over cubic, and possibly the first delay
>>> sensitive tcp that can compete effectively with it. I'm dying to test
>>> it thoroughly,
>>> but have a whole bunch other patches for wifi in my queue.
>>>> The authors suffer from typical naivete by thinking all flows are for file
>>>> transfer and that file transfer throughput is the right basic perspective,
>>>> rather than end to end latency/jitter due to sharing, and fair sharing
>>> While I agree *strongly* that lots of short flows is how the internet
>>> mostly operates, (I used to cite a paper on this a lot)
>>> a huge number of bulk flows exist that has been messing up the short
>>> flows. I think the number was something 70% of internet traffic has
>>> become netflix-like. *anything* e2e that can reduce the negative
>>> impact of the big fat flows on everything else is a win.
>>>> -----Original Message-----
>>>> From: "Jonathan Morton" <chromatix99 at gmail.com>
>>>> Sent: Sat, Sep 17, 2016 at 4:11 pm
>>>> To: "Maciej Soltysiak" <maciej at soltysiak.com>
>>>> Cc: "Maciej Soltysiak" <maciej at soltysiak.com>,
>>>> "cerowrt-devel at lists.bufferbloat.net" <cerowrt-devel at lists.bufferbloat.net>
>>>> Subject: Re: [Cerowrt-devel] BBR congestion control algorithm for TCP
>>>>> On 17 Sep, 2016, at 21:34, Maciej Soltysiak wrote:
>>>>> Cake and fq_codel work on all packets and aim to signal packet loss early to
>>>>> network stacks by dropping; BBR works on TCP and aims to prevent packet loss.
>>>> By dropping, *or* by ECN marking. The latter avoids packet loss.
>>>> - Jonathan Morton
>>>> Cerowrt-devel mailing list
>>>> Cerowrt-devel at lists.bufferbloat.net
>>>> Cerowrt-devel mailing list
>>>> Cerowrt-devel at lists.bufferbloat.net
>>> Dave Täht
>>> Let's go make home routers and wifi faster! With better software!
>> Dave Täht
>> Let's go make home routers and wifi faster! With better software!
> Cerowrt-devel mailing list
> Cerowrt-devel at lists.bufferbloat.net
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Cerowrt-devel