Port 8080 speed test

[Sean]

As I noticed some ISPs return considerably quicker speed tests with Ookla's speed than what's possible with regular web access including TestMy, I decided to snoop at how Speed test establishes its connection using Sysinternals' TCPView utility.  I sorted the traffic by 'Received Bytes' and then started a speed test.

 

While the multiple connections doesn't surprise me (Ookla's tests are all multi-threaded), what I was surprised with was what port it used - 8080:

 

 

It seemed like no matter what test server I tried, it ran its test over port 8080, which is a seldom used port for web traffic.  For example, HTTP and HTTPS traffic are carried over ports 80 and 443, respectively, while FTP traffic is carried over ports 20 and 21.   Port 8080 is typically used for an internal web proxy within corporate networks and for an ISP cache proxy in the early days of Internet for faster access to popular websites.

 

On the other hand, by running the speed tests over port 8080, this makes it easy for ISPs to prioritise traffic for anyone using Ookla's speed test as all they have to do is give elevated QoS for traffic running over port 8080.

 

I then thought - Is port 8080 necessary for Ookla's Speedtest?  To find out, I blocked port 8080 on my PC.

 

The speed test took a little longer to start, but once it did, it switched over to port 80, in this case with two threads:

 

 

Once I open up port 8080 and click 'Test Again', the next test runs on port 8080 again.

 

So an interesting idea would be if TestMy could add support for port 8080.  I don't think it will require much configuration other than configuring the various test servers to also accept traffic on port 8080.  Then it would just be a matter of typing testmy.net:8080 to perform the test on port 8080 with suspect ISPs, such as those that seem to throttle ports 80 (HTTP) and 443 (HTTPS).

[mudmanc4]

I find this disturbingly hilarious, not to mention in a sense, falsifying throughput details.

 

Hell, why don't we simply create a test on ou[r] local network to get 'interwebz' speeds. Oo

[CA3LE]

On 1/15/2016 at 2:05 PM, Sean said:

So an interesting idea would be if TestMy could add support for port 8080.  I don't think it will require much configuration other than configuring the various test servers to also accept traffic on port 8080.  Then it would just be a matter of typing [no longer works] to perform the test on port 8080 with suspect ISPs, such as those that seem to throttle ports 80 (HTTP) and 443 (HTTPS).

 

Thank you for the suggestion.  I added support for testing against port 8080.  As you imagined, simply visit [no longer works] -- works with all servers, similar to the https (port 443) test at https://testmy.net.

 

Let us know your findings.

[Sean]

It will probably be Wednesday before I get a chance to do some testing with the 4G networks.

 

With the Ookla App, I was able to force it to use port 80 with the help of the NoRoot Firewall App.  Basically I ran the speed test with all its ports opened (Speedtest used port 8080 like the browser test) and then repeated the test with only TCP ports 80 & 443 enabled.   Indeed there was a stark difference with the Vodafone 4G network, with the default test on the left and port 8080 blocked on the right:

 

 

While on my way to work, I stopped in an area where I get a strong 4G signal and used TestMy with HTTP vs HTTPS:

 

 

The Three network here seems to be throttling HTTP again, so there's also a stark difference between HTTP and HTTPS again:

 

 

So now I'm curious to repeat these tests with HTTP vs port 8080, especially with Vodafone which does its speed demonstrations using Ookla's App.  I suspect it will be faster than the HTTPS test above as it seems to throttle even HTTPS after about 20MB of data is transferred:

 

https://testmy.net/db/F8iglHK5a

Edited January 19, 2016 by CA3LE
screenshot to graphed

[Sean]

I did another test today in roughly the same 4G spot, this time comparing HTTP, HTTPS and port 8080.

 

HTTP vs HTTPS:

 

 

The port 8080 test was somewhere between HTTP and HTTPS:

 

 

With the Three network, the difference was not as great, but then again this time I ran these tests when I arrived at my workplace, so was about an hour later and a totally different location.

 

HTTP vs HTTPS:

 

 

Port 8080:

 

Edited January 20, 2016 by Sean

[Sean]

I asked my work colleague in the UK to try - the port 8080 bait made a juicy catch!

 

HTTP vs port 8080 - linear tests:

 

 

 

HTTP vs port 8080 - multithreaded tests:

 

 

 

HTTPS - linear:

 

Edited January 20, 2016 by Sean

[CA3LE]

Wow, that's a big difference on multithread for sure.  Next time you do those comparisons try to keep the test size the same.  Instead of that 6.8 MB and 42.1 MB ... maybe just do 50MB manual selection for both.  In that instance the 6.6 Mbps result would have taken about 60 seconds for a 50 MB test.  The 34.7 Mbps result would take about 12 seconds.  Takes a little longer but it's a better comparison, apples to apples.  Don't get me wrong, the results you show above are a great comparison.  I'm just pointing out how to make them a little more definitive.

[Sean]

So far I haven't heard back from my work colleague and am sure he is wondering why I keep asking him to run more and more tests on his connection.

 

However, by looking at previous results of his user ID, I got the previous multi-threaded test I ran a few weeks ago and it had a 51MB block.  That was the one I remember that made it difficult to speak to him over VoIP until the test completed.

 

 

The following is a better example from the Three mobile network on 4G, where I chose two equal size blocks for both up and down - http vs port 8080:

 

 

My workplace connection does not discriminate between http and port 8080.  There was no one else in at the time, so had the connection to myself:

 

[CA3LE]

Well, now isn't that something.  Tricky tricky.  That's pretty definitive proof.  Really the only other probable way you could get a result that dramatic when you keep the variables that level is if you purposely throttled back your own connection.  And if they're doing it, you know that others do as well.  We need more proof across a wider demographic.

 

I'm creating an addition per another users suggestion that just came in.  I think it will work great to build groups of users that want to build a claim like this.    I also need to build in a way to quickly differentiate the port 8080 and https tests in the results. 

 

Purposely allowing ports that they know speedtest/ookla uses to be faster and throttling back others.  What a joke.  Has to be borderline illegal in some way, right?  I really hope that more people start finding TMN so they can know the real deal.

 

Thank you for continued support.  I really appreciate your comparisons and suggestions to help detect this scam deeper, you're making TMN a better tool.

[mudmanc4]

I'm looking for any reason for this other than specifically,  deception, anyone?

 

Again, unless the test is taken and strictly meant to be used by the end customer, and only to be used in a manner to verify ISP to end user connectivity.

[CA3LE]

8 hours ago, CA3LE said:

I'm creating an addition per another users suggestion that just came in.  I think it will work great to build groups of users that want to build a claim like this.

 

https://testmy.net/group

 

Here's an invite: https://testmy.net/invite/awesome/lAoun43

[Sean]

I spoke to my work colleague last Friday that I was going to write an article about ISPs inflating their speedtest.net results.  So he ran two 50MB multithread tests this time without any other network activity, i.e. no remote desktop or VoIP call traffic either.  The port 8080 multithread test is on the right. 

 

 

Let's hope this article grabs some attention. 

[CA3LE]

1 hour ago, Sean said:

Let's hope this article grabs some attention. 

 

Wow.  What a great read, I put that on the homepage.

 

You're really a great writer, don't think I've told you that yet but even before now I've been very impressed.  This just sets it over the top... way over the top.  Thank you.

[CA3LE]

@mudmanc4 brought this to my attention.

 

http://club.myce.com/f186/internet-providers-caught-inflating-speed-test-results-340534/#post2767288

 

Can you explain to them that TMN represents some of the most common connections, in the most popular hosting areas in the world.  And all of the testing locations have a minimum of 1000 Mbps uplink.  These datacenters are very popular, well connected and are regularly tested for quality.  Speedtest.net has people believing that you need to test off a server that's close to you.  In my opinion testing close does provide good information but only when compared to a test further away... 1000 or more miles away can EASILY provide great results IF the provider is delivering.  I do it EVERY day.

[Sean]

As the Three mobile network here still gives traffic priority over both HTTPS and port 8080, I tried some experimenting with HMA's VPN to see if I can exploit it.  I never thought of bringing my laptop at work, so was limited to running the HMA App directly on the phone.

 

After about 4pm, Three was giving a fairly steady 2.0Mbps to 2.5Mbps over HTTP and 4.5Mbps to 5Mbps over both HTTPS and port 8080 based on a series of TestMy tests.  The following was an example about 4:30pm, HTTP vs Port 8080:

 

 

I established a VPN connection to the HMA's Irish server and got about 3.5Mbps with the plain HTTP test.  Based on past testing with HMA, its London server for some reason has lower latency than its Irish server, so connected to London and got a HTTPS-like speed test with the plain HTTP test.

 

 

With multi-threaded tests over port 8080 (no VPN connection), I was able to get between 10Mbps to 12Mbps, 4 x 50MB tests in a row. The Speedtest App for some reason couldn't hit 10Mbps even after a few runs using Three's own Speedtest server.

 

 

While privacy concious people generally avoid HMA due to its logging policy, the main reason I use it is for its flexibility of connecting to its servers, such as over UDP and even different port numbers.  So if I remember to take my laptop the next day I'm in and get a chance, I'll try to see if there's a way of to exploit any extra bandwidth.  I remember being able to get around 2Mbps from a 512Kbps limited hotel Wi-Fi connection using a UDP connection.  The Android VPN functionality is fairly limited, e.g. fixed port, no UDP support, etc.

[Sean]

Over the past week or so, I'm no longer able to access TestMy over port 8080.  HTTPS works fine.

 

As of the 30th April 2016, the EU rules on net neutrality came into force.  This means that all that traffic must be treated equally regardless of the source and destination and that ISPs are no longer permitted to block and throttle traffic. 

 

In previous tests I carried out on the Irish Three mobile network (mobile data on Smartphones) and the UK ISP Pulse8 broadband, there was a vast difference between the throughput I got over normal HTTP compared with port 8080 that Ookla's Speedtest uses. 

 

If port 8080 can be enabled again on TestMy, I'll have a go at making a video capture of running tests on TestMy compared to accessing it over port 8008 as well as two third party tests on my phone with the Three network, e.g. Leaseweb test file (normal HTTP download) against running Speedtest.net (which uses port 8080).  If there's a clear variation, I'll put it on YouTube. 

Edited May 1, 2016 by Sean

[CA3LE]

Some automatic action must have rewritten http.conf on the main server.  This has been corrected and should now work for you again.  Let us know what you find.

Edited June 14, 2018 by CA3LE
port 8080 experiment is now disabled

[Sean]

Thanks for that - Indeed there's still a clear difference between http and port 8080 on the Irish Three network, at least on a Smartphone.

 

Three 4G vs Meteor 3G - Linear test with 6MB block:

 

 

Three 4G vs Meteor 3G - Repeat test over port 8080:

 

 

Three 4G vs Meteor 3G - Multithread test with 50MB block over port 8080:

 

 

I've a video recording captured.  If it recorded fine, I'll trying editing something to post.  I also tested with Ookla Speedtest and Leaseweb's test download and got similar results, i.e. Speedtest was considerably quicker on Three as expected but Leaseweb's test file downloaded far quicker on Meteor 3G.

[Sean]

It's Showtime...

 

Part 1 - Browsing the web and a 10MB download on Three (4G) and Meteor (3G):

 

 

Part 2 - So what does Ookla Speedtest report?  Can TestMy be run in a way to match it?  Let's find out...

 

 

[CA3LE]

Sweet videos!