Where does internet speed get measured from?

rpc

Hi. Apologies, but a bit of a trivial question: where does the Fing app take its speed measurement from?
I'd assumed that it would be at the Fingbox on the basis that the point of the device is to monitor my Internet connection and not my internal LAN. But the speeds reported are: desktop is ~800Mbps (download) and ~49Mbps (upload); mobile phone is ~30Mbps (down) and ~44Mbps (up); my laptop is in between these measures.
It seems, therefore, that the speed is being measured to/from the device that the app is on - and they will all potentially throttle the results to some degree. So, in order to know the actual speed delivered by my ISP it has to measured at the router itself and the Fingbox is irrelevant to this.
I'd appreciate confirmation or correction of the above. Thank you.

Chris_MLAB

I suspect that these differences are due to the NDT client library being used to conduct the test. But this leads to more questions as well. In the case of our website, the client code is in JavaScript using WebSockets.

@Fing team:

• what client code is being used in the Fing mobile app?
  
• what client code is used on the Fing box?
• what system architecture is used on the Fing box?
• are tests from either the Fing App or Fing box conducted over TLS? or are they plain text?
  

If the Fing Box uses an arm system architecture, and the NDT test defaults to TLS, then the overhead of encryption may be affecting the ability of the system to measure effectively. I've confirmed this personally using armv7 systems, but not yet on arm64.
Another thought would be to check on the NDT client library being used on Fing box. Our most current NDT reference clients for systems integrations like Fing box are in Go [1] [2] but there are older NDT clients that used other languages.

I hope this is helpful. @Fing please feel free to reach out to M-Lab directly to coordinate follow up if needed.

RayW

I have Fing Desktop 1.1.0 installed on each of 3 Windows 10 machines.  One of the installations consistently shows much lower Internet Provider speed test results with Fing than with others, e.g. Ookla.  The other two do not.  All three consistently show the same Ookla Speed Test results.

It seems to me that Fing must be 'seeing' something unique about that particular system, but I can't be sure of what it must be.  It seems likely that it is something in the network stack - but it seems curios that I've found no other speed test utility that is affected by it.  Wireless connectivity is disabled on all three machines.  All are running the latest Windows 10 updates.  All are connected via ethernet cable to the same switch.

Maybe Fing is getting too intimate with the hardware.  Each of the three machines has different ethernet hardware:

1. Intel Ethernet Connection (2) I219-V
2. Qualcomm Atheros AR8171/8175 PCI-E Gigabit Ethernet Controller (NDIS 6.30)
3. Realtek PCIe GBE Family Controller

The machine demonstrating the issue has the Realtek hardware.

I am curious to know what the problem is.  If there are such hardware dependencies, I'm not yet willing to trust Fing Desktop as a go-to diagnostic tool.

kltaylor

Hi @RayW , welcome to the community and thanks for the question.

Each speed test variant that is available on the web all step from different servers located in different locations.  The geolocation for the servers alone will cause you to have a variety of different results.

Another reason would be in the ownership of the server and service.  As an example, Speedtest (arguably one of the better-known testing sites) is owned by Comcast.  In which case according to them, I have almost a 200 MBps download rate, but the site that I consistently use (testmy.net) tells me otherwise.

My advise is to find a service that is close, one that you like and stick with that as your baseline.

RayW

This doesn't my real question.  I suppose it's my mistake for not having enough detail in the question statement.  I was too impatient to spend the effort to condense all the pertinent context into the limited allotted space.  In retrospect, I do see that the base question, as currently worded, begs the obvious answer centered around the choice of speed test service - and that such would explain some reasonable variation in test results.  I had hoped that the context info I had included in the more generous input area would have gotten equal consideration.

The real problem is that only one of three machines configured and connected alike demonstrates very substantial differences in measurement results, e.g., 40 Mbps instead of the expected 200+ Mbps.

RayW

Here's more info.  I ran some tests with the site that ktaylor uses, testmy.net.  It produced pretty much the same results that Fing did, i.e., two of the three machines showed over 200 Mbps (consistent with both Ookla and Fing) and the third machine showed around 15 Mbps (consistent with Fing, while Ookla showed over 200Mbps).

kltaylor

@RayW:

Thank you for more clarification ... in this case I'd look more closely at the device that's not providing a better speed result than the others.

Is that device wireless?  Is it older hardware where maybe the network adapter is a 10/100 and not a 10/100/1000 card?  Check the Ethernet cable too if its connected via that, it could be that the cable itself isn't Cat5 or Cat6. 

I'd open Task and Resource Manager to investigate the network controller device closer.

RayW

I'm now using Fing 1.3.1.  I'm still seeing the aforementioned problem.

It turns out that if I watch Ethernet Throughput using Windows 10 Task Manager, I see exactly the same throughput readings as indicated by the application currently performing the speedtest.  So, for example, when running the download speedtests sequentially, back-to-back on the machine demonstrating the issue, testmy.net shows 10.4Mbps, fing shows 39.6Mbps, and Ookla shows 230.4Mbps, each reading corroborated by the Task Manager.  In the past, upload values across these speedtest apps have been fairly consistent at around 36Mbps -- however, lately during upload tests, testmy.net responds with "Something is wrong... The webmaster has been notified of this error." on all of my machines, i.e., with or without the issue being discussed.

I don't think it is a problem with the local machine hardware or OS, but more a question of how the speedtest app is performing the test.  I wonder how Ookla and the other prominent speedtest apps manage to get consistent speed test results across all machines.

Robin_Ex_Fing

Hi @RayW
Can you please check if M-Lab speed test is the same from link below?
https://speed.measurementlab.net/#/

That is the partner we use for speed measurement, so if Fing results are not in line with the ones reported by M-Lab, we definitely have to fix something.
Also, please check the Accepted answer on this post for better information.

RayW

Tests on a given PC were nearly concurrent.  But one PC was tested a few hours after the other.

               PC with Issue                        PC without issue

               Down              Up                 Down            Up
MLAB     38.23Mb/s      31.8Mb/s       194.09Mb/s    36.58Mb/s
Fing        38.0Mb/s         33.3Mb/s      229.6Mb/s     37.9Mb/s
Ookla      225.99Mb/s    36.86Mb/s     229.82Mb/s   37.9Mb/s

Robin_Ex_Fing

Hi @RayW
As I can see both Mlab and Fing results are consistent so i believe either Mlab are facing some server overload in these times.

We have been in contact with our Speedtest partners, M-Labs. They confirmed that they have recently rolled out an upgraded of their infrastructure & migration of machines to a new platform. This is being done to improve the overall Speedtest performance going forward. By doing this, initially they have cut of the multi-stream download Speedtest that we rely on for our Speedtests. However, they have greatly improved the single stream download Speedtest (previously the single stream download could measure up to 250 Mbps, where now it can measure up to 500+ Mbps). With speeds above this, the measured speed reported will be lower than before, due to the single stream limits.

To note for Fingbox users: this will also affect Fingbox Speedtests. We have released a new firmware version for the Fingbox which will allow single stream download to work again, firmware version 1.14.13 and onwards. If you have not yet received the new release, please power-cycle the Fingbox, this will automatically update the firmware on the Fingbox.

M-Labs are working on re-introducing the mutli speed download speedtest, to include it again in the future. Currently, we have no time-frame on when this will be re-introduced. Unfortunately at this stage, there is no further steps we can take to expedite this process.

Marc

@Robin or @Ciaran, could you answer @rpc question?  Thanks...

Robin_Ex_Fing

Thanks @Marc
@rpc
Yes, Speedtest is measured from Fingbox. Please check this link for difference in speed test results: https://help.fing.com/knowledge-base/internet-speed/

Marc

@Robin if this helps, here is the Speedtest results from 3 different devices.  We have a Fingbox, Desktop via mLabs and Desktop via Fingapp.  Desktop and mlabs seem to be the same, Fingbox not so much...  Its always been this way by the way.  Fingbox always reads slower then mlabs.

Beantown

Hi - I see the same when using my Fing App (iOS) to test internet speed. When I use speedtest app directly I get much faster results. When I went to the mlabs site I also get faster results. Is there a fix for the App/Fingbox?

Wintveld54

The speedtest on the fing desktop is significant higher (480 Mbps on a 500 Mbps connection) than the speedtest on the fingbox (208 Mbps) It is not occasionaly but over a long period of time (weeks). Is this normal or is something not OK

Robin_Ex_Fing

Thanks @Marc
I am tagging @Chris_MLAB if he can spread some new information on this. Any advice @Chris_MLAB? Thanks

Robin_Ex_Fing

Thanks @Chris_MLAB for the valuable information. I am seeking further advice on the questions you asked. I will update everyone shortly.

Robin_Ex_Fing

Hi All, I have merged few tickets for the Speed test issue. Please check Chris_MLAB reply and here are the answers for the questions he raised. Tagging @Marc for your visibility.

• what client code is being used in the Fing mobile app? NDT5
• what client code is used on the Fing box? NDT5
• what system architecture is used on the Fing box? ARMv7 Cortex

we are using NDT5 protocol (not NDT7 as their web client). The tests are conducted over plain socket -> no support for TLS