[Cerowrt-devel] bufferbloat and the web service providers
dave.taht at gmail.com
Sun Dec 9 11:36:01 EST 2012
The results of the rrul benchmark in conjunction with the chrome web
page benchmark tests in combination with the debloat and simple_qos.sh
script in combination with linux 3.6.9 are so frigging spectacular at
4Mbit uplinks and above, that I'm reluctant to publish them.
I've tested dozens of web sites ranging from google to xfinity at this
point with improvements in web page network load time ranging from 4
to 10x under otherwise saturating workloads. Another way to look at it
is that most web page loads I've tried degrade by less than 20% while
under the extreme stress of the rrul benchmark. (vs 10x worse or not
completing at all under pfifo_fast or short queues)
Others have had great difficulty in reproducing these results, so I'm
gradually trying to reduce the sources of experimental error and
improve the scripts and documentation and so on... Recently I got bit,
bad, by the recent default of the GRO offload to a couple network
drivers.... my current simple_qos script only uses ecn on ingress...
So I'm setting up 3 labs to nail all that down and also work on
artificially induced longer RTTs.... which is going to take some time.
I hope to make available a few netperf servers for public use fairly
soon (or line up some volunteers to do so?)
In the meantime I'd urge folk to track the development of the rrul and
related benchmarks, install their own netperf (netserver) servers, and
Upload speeds below 4Mbit remain a bit problematic and wifi a nightmare.
The 3.6.9-5 release of cerowrt can run enough netperf servers to keep
100Mbit links busy but only just barely, and in some tests, we overrun
the switch's buffers, not fq_codel, so htb is the way to go, to
another box, through the router....
On Sun, Dec 9, 2012 at 5:14 PM, Maciej Soltysiak <maciej at soltysiak.com> wrote:
> What are the heaviest (amount of elements, css, images, scripts, js bugs, ad
> trackers, all that filth) websites out there?
> I wonder if it'd be worthwhile to demo loading that in a standard bloated
> environment and compare with debloated setup?
> Maybe that could be a variant of your idea below but geared towards end
> On Mon, Nov 19, 2012 at 10:28 AM, Dave Taht <dave.taht at gmail.com> wrote:
>> I am not into politics. Really. A problem, though is getting the NN
>> people on both sides to stand down, because the real problem was
>> bufferbloat - and then work together to move forward, on an
>> engineering, rather than political basis.
>> So, incidentally, while I'm suddenly paying attention to politics,
>> while trying to find some way to get this engineering effort funded
>> for next year... I noticed a new lobbying group has come out of
>> stealth mode, representing the web apps people, google, facebook,
>> salesforce.com etc
>> Anybody know these guys? I have long hoped to get these services aware
>> that they needed to help fix bufferbloat if they wanted their cloud
>> based businesses to work better.
>> IF I felt like expending money on a marketing campaign to explain our
>> issue, and get some funds towards fixing wifi in particular -
>> the script to the video would look like this (leveraging the old
>> "brain on drugs" meme)
>> * This is your cloud based business - showing a fast, normal load time
>> of some major website
>> * This is your cloud based business on bufferbloat - showing a loaded
>> webload go to hell of that site
>> * This is your cloud based business with bufferbloat fixed -
>> * Any questions? - point back to campaign site
>> Then create a copy of that 30 second piece for every one of the web
>> service providers listed as supporting the above org. And as many
>> others as possible. And get it on youtube. And buy some ads.
>> I have no idea how much that would cost to produce professionally (?)
>> (at a cut above what jg has already done), but I'm certain it would do
>> some good, both in raising awareness and maybe gaining funding for
>> next year.
>> Dave Täht
>> Fixing bufferbloat with cerowrt:
>> Cerowrt-devel mailing list
>> Cerowrt-devel at lists.bufferbloat.net
Fixing bufferbloat with cerowrt: http://www.teklibre.com/cerowrt/subscribe.html
More information about the Cerowrt-devel