Re: [Bloat] new public web tests for bufferbloat

[Dave Taht <dave.taht@gmail.com>]

On Wed, Jun 1, 2016 at 4:06 PM, Alan Jenkins
<alan.christopher.jenkins@gmail.com> wrote:
> On 02/06/2016, Alan Jenkins <alan.christopher.jenkins@gmail.com> wrote:
>> On 01/06/2016, Alan Jenkins <alan.christopher.jenkins@gmail.com> wrote:
>>> On 01/06/2016, Dave Taht <dave.taht@gmail.com> wrote:
>>>> see: http://labs.thinkbroadband.com/speedtest/
>>>>
>>>> (can't test myself, not being in england - can someone there test it
>>>> and post results/screenshots?)
>>>
>>> http://www.thinkbroadband.com/speedtest/results.html?id=1464821570474383655
>>>
>>> (OpenWrt SQM. ISP _download_ is un-bloated already, for reasonable
>>> numbers of streams. Though even with SQM, downloading multiple
>>> torrents in Transmission can cause both latency up to 100ms & packet
>>> loss).

I have generally felt that torrents would be better managed by the
ISP's use of tools like fq_codel or cake on their rate limiters,
rather than managed on the cpe.

a note from my nicaraguan trip - which I will write up more fully
later - was that the torrent problem had essentially vanished there in
the 5 years I was gone. The cybercafes had dropped from 6 overflowing
ones to two (as all the hotels had got their own links), and most
tourist and local usage had moved to smartphones *everywhere* rather
than laptops.

One former cybercafe owner I interviewed said "all they know is what
the phone offers" - and "skype and facetime" were all they wanted in
addition to web, email, snapchat, whatsapp, and youtube - they didn't
know what a torrent was. I found a surprising number of people using
whatsapp....

Latencies over the network were often horrible, yet several people
noted to me, while talking about it, how much better the wifi was
there than in the states. My channel scans showed fewer than 6APs
"hearable" everywhere I went (in multiple cases less than 2), and
there was little to no 5ghz present.

Both the "main" networks I got access to were completely behind
private ip address space - the condor.co.ni WISP (in one case) had 3
hops to the next public ip, the cable ISP - *6*. Sigh, an entire
country almost entirely behind NAT...

I left 2 archer c7v2's behind on those networks to see if they help...
and can survive the heat and humidity. The 5 mbit/.5mbit cable link I
left one on had over a second of download latency under load, 3
seconds up, before applying sqm - and one WISP link (1.5Mbits
symmetric) - over 600ms. Another WISP link - well over 3 seconds
sometimes but the problem was often 1-3 hops into their network....

>>
>> oops...
>>
>> the results page above omits the bufferbloat result.

Sigh. Can bugs be filed?


> The original
>> result has a sharing button that requires a Facebook login.  Here's
>> the screenshot

what was their bufferbloated result?

>>
>> https://dl.dropboxusercontent.com/u/49925445/bufferbloat.net/thinkbroadband.com/2016-06-01%20labs.thinkbroadband.com%20speedtest.png
>
> Sorry for noise, I should also say:
>
> The latency: 58ms is due to testing on wifi (I guess it's the minimum
> over a number of pings).  The "A" grades for bufferbloat do not seem
> expandable / clickable
>
>>>> and:
>>>>
>>>> https://sourceforge.net/speedtest/
>>>>
>>>> which is quite pretty, if mildly confusing.
>>>>
>>>>
>>>> I have issues with the language and testing methodologies, of course,
>>>> but it is good to see these arrive. When will speedtest.net get it
>>>> right?
>>
>
>
> --
> A: Because it messes up the order in which people normally read text.
>> Q: Why is top-posting such a bad thing?
>>> A: Top-posting.
>>>> Q: What is the most annoying thing in e-mail?



-- 
Dave Täht
Let's go make home routers and wifi faster! With better software!
http://blog.cerowrt.org