Speed tests and the missing megabits: why you can't hit 1Gbit/s

Internet users might remember the days of dial-up connections and the excitement of looking forward to a noticeable difference in experience every time the line speeds were upgraded. Today, however, customers don’t seem to feel the same difference when upgrading. This is where it becomes necessary to understand the differences between capacity and throughput, and to explore evolving connectivity requirements.

A favourite thing for people to do when they get an internet connection or upgrade is to run a speed test. But something strange happens: a customer with a 500Mbit/s connection might download a file from a fast server and achieve 480Mbit/s; when they upgrade to 1 Gbps, they might expect to see 960Mbit/s, but instead, they only get 600Mbit/s.

External factors impacting performance

At lower speeds, the internet connection itself is the bottleneck – meaning that you can get close to the plan’s advertised speed. But beyond certain speeds the bottleneck is no longer your access speed, but rather external factors such as servers, end-to-end network conditions and the TCP/IP protocol.

So, even if you have a 1Gbit/s, 10Gbit/s or 100Gbit/s connection, if the maximum throughput achievable due to server-side or network limitations is 600Mbit/s, that is all you’ll get. In this example the capacity is 1Gbit/s while the throughput is 600Mbit/s. As fibre internet access line capacities continue to increase, these limitations will become more noticeable.

The same applies to wireless services, where, for example, even though 5G may be capable of a theoretical speed of 20Gbit/s, this is based on using the best equipment in ideal conditions, and even then there are radio frequency limitations that will prevent this speed from being achieved.

Internet service providers also make use of contention, which segregates different service profiles. A dedicated service will have much less contention than a broadband service. Higher-contended products allow customers to use their full connection simultaneously, and therefore customers share a portion of the network’s capacity. In such instances, users may observe higher throughput at off-peak times, and lower throughput during peak times.

Speed tests not an accurate picture

Back to speed tests: to be accurate, the speed test must be done in a manner that removes all other variables, meaning using a wired connection directly from a laptop to the router. Users might have to make sure the device being used for the test is capable – if a laptop has a 100Mbit/s network port, that is the maximum you are going to get. Even the quality of the LAN cable may influence your performance test.

Servers, switches, routers, cables, firewalls and access points can all have a negative impact on speed. For example, when fibre network operators were rolling out free speed upgrades in recent years, customers found they were unable to benefit because routers were incapable of handling over 100Mbit/s. Many were unaware of this and thought the problem lay with the fibre provider or ISP.

The second challenge is that high speed connections, such as a gigabit link, were not designed to deliver 1Gbit/s to a single user on a single device but rather to connect multiple users, devices and applications concurrently to the same network. Having a gigabit connection is not only going to ensure that a single user has perfect video calls all the time; rather it means that multiple users, all taking part in video calls at the same time each have an optimal experience.

The combined required throughput of a link can be determined by the simultaneous use per user or device. As example, should 30 concurrent users/devices require an effective throughput of 10Mbit/s each, one would then need a 300Mbit/s service. It is all about ensuring that each user/device has a reasonable experience. Here, the limitations of the internet protocol or devices no longer apply, because it’s not one device trying to download a file at 1Gbit/s, but multiple devices that are accessing cloud-based services, downloading, streaming video and gaming online at the same time, and making full use of the bandwidth available.

More to connectivity than just speed

We’re starting to see that a speed test is no longer an accurate reflection of what you can do with a high-speed internet connection – to run a test properly for the modern use case would require you to fire up multiple connections concurrently and test the total capacity of the connection.

We grew up in an age where there were severe limitations on local networks and this was the bottleneck; with such low speeds, every bump up had a noticeable difference like significantly faster download speeds. It also meant that speed tests had more relevance back then. However, as technology has evolved this is no longer the case and doubling your line speed is not going to result in being able to download a file in half the time that it used to before.

On fibre services the line speed is the maximum throughput possible of the line. On wireless connectivity one has a practical achievable speed based on network load and a theoretical speed that will never be achieved in practice. We briefly mentioned theoretical and real-world speeds in relation to wireless services, but as we start seeing higher speeds on fibre lines, it is likely that we are also reaching the theoretical speed through that medium as communication as well – for example, how can you properly test a 10Gbit/s line when the devices themselves are not capable of handling such speeds due to limitations in processing power, memory and other components?

Reality: we are reaching a stage of bandwidth abundance, where service providers can provide users more bandwidth than what they need. Here, speed is no longer everything, and what is important is having the capacity to ensure a quality experience across all users and devices.

The authors are Theo van Zyl, head of wireless at Vox, and Andre Eksteen, senior product manager of fibre to the business, also at Vox
Read more articles by Vox on TechCentral
This promoted content was paid for by the party concerned