Our Blog

Latest news and updates from Baltic Broadband

Why We Think Speed Test Sites are Fundamentally Flawed

Posted by matt.wilson on 16 01 2018. 0 Comments

We often get asked by customers as to why speeds vary across the various website speed tests and also in some cases misreport the actual speeds being delivered to a customer.

There are many reasons why such sites are not as accurate as one may think. Speed Test sites are usually hosted by volunteers who use “spare” bandwidth on their network in return for free advertising or to help their own customers achieve better speed results.

Speed test sites were really on designed to give an indicative idea and not conclusive answer and are volunteered by other people with no guarantee or SLA as to their capability, so therefore cannot be relied on. It’s not our intention to compete against a volunteer hosting a website in return for adverting money.

How it Works

All the speed-test website’s actually do is download a small set of files from a remote server and provide a number or dial on the screen as a “guesstimate”, but it doesn’t have any scientific calculations behind it.  By doing the test, your not actually testing your internet speed, your performing a test to a specific server somewhere on the internet, a place in which you’ll probably never visit again, i.e as part of your normal usage.

Here is a breakdown of what can affect your reported speed:

The Speed Test Server Itself

Quite often the actual speed test servers are hosted in a virtualised environment, meaning that the owners are running many servers on a single physical computer.  During busy times, the Speed Test server itself may actually be too busy (serving other speed tests to other users) to serve you in any reliable way and can produce inaccurate results.

The Speed Test Server’s Connection

The connection between you and the Speed Test server might not be capable of measuring the speed you require, for example, the servers bandwidth might already be used up providing other Speed Test measurements to others.

In this case, the Speed Test server simply does not have the bandwidth in the first place to measure your speeds.  Quite often the server is connected on a 1Gbps or 10Gbps connection, but if there are 50 other people using it at the same time, how can it actually test you at say 350 Mbps if it does not have it to give?

Think of a car race whereby your car does 200 mph but the one measuring you can only do 100 mph, once you go over 100 Mph, how can the testing car say you did or didn’t manage to get to 200 mph?  Many volunteers who host the speed test servers don’t have dedicated bandwidth set aside to actually deliver the bandwidth you need for that test, as quite often volunteers tend to use spare bandwidth that is not being used at that time, a concept which of course is fundamentally flawed.

We found that www.thinkbroadband.com/speedtest it more accurate than some,  but when we tested it, it was capped at 90Mbps download and 150Mbps upload, but totally useless for 100Mbps+ lines. It also had high latency, and was running over SSL so had those overheads too which degrade speeds.

Many speed test sites were designed for consumer grade ADSL (asymmetrical) – not wide band and fully synchronous lines.  We generally find that when you try to test anything over 100MB, many sites bottom out and say its only 100MB when it could be even 500Mb that you have.  This is because many providers have either throttled back the amount of “free” bandwidth that they are using for their servers, or have decided only to support up the likes of 100MB as that is the highest speed that the provider currently supports to their customers.

A good example is BT’s own Broadband Performance Test, in which even they say it cannot be tested for connections above 330MB.

The Speed Test Websites Themselves

Some sites runs on the outdated Adobe Flash, and some without Flash. Firefox doesn’t even support Flash anymore, that tells you how outdated that technology is. Well, that also affects your speed test results.  You can run tests yourself on both and see the huge difference in reported speeds, all by changing out Adobe Flash.

Anti-competitive Measures

We now often see ISP’s deliberately slowing down traffic from a competitor in a bid to make themselves look better.  Think about the VW emissions scandal and how they created “defeat devices” in order to fool the tests. Some ISP’s have been making themselves look good on the Speed Test websites by downgrading traffic from competitors (like us) in order to generate more sales. Unfortunately its the consumer that loses out as this fake presentation of speed only serves the gaining ISP. Of course we feel these tricks are dirty tactics and dishonest and need to be disallowed in our industry.

The way we connect our customers to the Internet

The way we connect to other ISP’s and networks around the world doesn’t help, we connect and “peer” at internet exchanges in the core of the internet.  While this gives a really fast and unfiltered connection for our customers, some Speed Test servers are hosted on the outside of an ISP’s network and not at the core, behind many firewalls and appliances, again having the possibility of affecting the reported speed.

Your Computer

Background programs such as anti-virus software, corporate VPNs, peer2peer clients etc can all slow down the connection unexpectedly and therefore report to you inaccurate results.

We’ve seen a number of tests done by older computers, that have a low amount of memory and processing power to render the imagines inside the browser.  One customer who contacted us about reported slow speeds even had multiple browsers open, of which once all closed, reported a 240% increase in speed (as the computer had more memory to handle the test!)

So…That’s not all, but I think we’ve made our point..!

Our Tests

We’ve done a number of tests, both inside and outside of our network, including building our own speed test server, complete with a 10 Gpbs network card, and even when we did, we found varying results, many of which did not relate to the actual LAN speed.  If even we cannot get it giving us a full accurate speed when we were sat next to the server, what hope do customers have when they could be miles away?

In one basic test, we purchased 3 separate laptops from ebay.  One very old, one 3 years old and another one brand new.  All were connected to the same connection and ran one after each other.  The speed tests showed almost a 52% difference in speeds between them, with the newest of laptops showing 210% speed increase.

 

The Answer and Conclusion 

There is no real answer or testing website that can help unfortunately.  We don’t believe that the science of the transmission of data can be simply measured using a website.  While the speed test tool websites can be useful, especially for ADSL and home users and the testing of lower speeds (<50Mbbps), it is on our experience not an exact science and more of an art.

Instead, we believe that the the answer is for the ISP to perform a test from the customers router (CPE) to a series of endpoints or peering points on their network (for example using the likes of iperf) and take some averages (but even then, its only reporting the experience to a handful of commonly used websites,) but it should be able to determine the speed from your premises to the point in which your ISP connects to the Internet, meaning that they did their job correctly and the rest is up to the rest of the Internet.

The fact is, the very nature of the internet and how it works creates this problem.  Though your use of the Internet, you are connecting to many servers around the world, each with their own workload, capability and fed from different internet connections themselves, some are throttled and are some not,  and the true fact is that your speed to such servers is different each time you access them as well as the route your traffic takes from your device to that server, which could change from one minute to the next. It’s a bit like playing a game of football whereby the goal posts are moving every few seconds, all around the pitch and faster than you can keep up!

Is all lost?

Not really, we always say that a customer should rate the speed based not on a dial or numbers on a website, but on their overall speed when using the Internet, like the load-times of websites, watching video streaming and the downloading of files.  We think that the overall feeling and experience is worth more than a Speed Test representation of what it thinks you are getting.

What we feel is important is not the speed that one single user gets or can get by doing random speed tests but is having enough bandwidth for everyone and every application when everything is running at the same time – inside your business.

We feel that it’s not the speed that counts, it’s what you do with it and the experience the end user has at the end of the line.

A Technical Test

A good guide, again not conclusive is to download a large file as below and see how long it takes. This is a better way as it does come from a trusted source that is known to have a large and capable connection:

  1. Go to the HSO Linx Website and select one of the distributions of your choice, as big as file as possible..
  2. Watch the meter on the top right, only pressing download when it is very low
  3. Note the speed quoted on the download
  4. Take the max speed shown in your browser and multiply it by 8 (because of 8 bits in a byte)
  5. You should have a good idea of the current speed of your connection assuming everything is working correctly but again, as so many things can affect speeds, it is only a rough guide

Other Reading

It appears that this is quite a hot topic currently discussed by many others, read more at the links below.

http://support.blazingseollc.com/support/solutions/articles/19000045815-speedtest-net-is-wrong

https://community.plus.net/t5/Fibre-Broadband/Ookla-Speedtest-Strange-results/td-p/1416643

 

 

Leave a Reply

Your email address will not be published.