ISP Speed Trend
Speed Trends (by technology and tests)
Trend of Average Website Speed
TrueNet measures the average speed of some 30 webpages, every hour of every day. Measurements are every hour of every day from all panelist connections, we have approximately 400 panelists. ISPs cannot influence maximum webpage speed, that is due to the upload speed of the website. However, line congestion and technology can and does influence speed at some times during the day (ToD), usually 9pm when most people are busy on the internet. For ISP comparisons we show that time, for technology or product comparisons, we show the average speed at all times.
The best ISP consistently has the highest minimum hour for the month, ie they consistently provide a speed during the evening busy hour that is closest to the fastest the line will allow. We experience many complaints from panelists of unusable performance if the average drops below about 75%, so we consider 70% the natural zero, while some users find performance is impacted if performance falls below our 90% target.
We have also report different technologies, because the results are surprising, Fibre performance does not significantly improve for speeds higher than 100Mb/s, and due to the early supply of 1000Mb/s or GigE services in Dunedin, we see that product take time to even get as quick as 100Mb/s.
The technology comparison is as a percentage of Fibre speeds, this is because the pages change from month to month, so absolute speed is not that relevant. We index our charts to Fibre 100Mb/s becuase that is the most stable, with a typical average of 30Mb/s for the webpages we test.. Chart 1 shows:
- Fixed Wireless is slowly decreasing in speed as demand rises.
- Cable, while mostly 100Mb/s like Fibre is slowing and less than 60% of the speed of fibre
- Cable and VDSL are pretty similar in performance
- Spark & Skinny Fixed Wireless are about the same as ADSL
We do not compare peak speed of 1000Mbps and 100Mbps, partly because test equipment and line speeds are not yet possible for 1000Mbps, but also because it is just not that important. Most activities on the internet with very fast conections are dependent on the supplier, a supplier with only 100Mb/s connection is not going to even deliver 100Mb/s, let alone 1000Mb/s. The most critical activities are surfing and video, as you can see on chart 2, the fastest surfing speed is under 35Mb/s, most video is limited in speed by the supplier, Netflix about 5Mb/s, Youtube about 20Mb/s and slowing over time. There is a good reason for video limitations, most video watched uses less than 5Mb/s to view, so faster downloads only reduce the probability of buffering, they do not make the video clearer, only very high quality will make this happen and that is not yet common.
Chart 2 shows:
- 1000Mb/s fibre was only as quick as VDSL until late last year - simply because early connections were in Dunedin, where distance to the service is an issue
- 30Mb/s fibre is getting slower - possibly due to congestion. However, the connection is subsidised, so now is the time to get fibre, you can always upgrade to higher speeds later
- All Fast fibre (1000Mb/s, GigE, 200Mb/s and 100Mb/s) download our sample pages at the same speeds.
Comparing ISPs we see that some are better than average while some are less - as expected. The dashed lines are for those with fewer, but sufficient panelist samples.Chart 3: Average website speed trend Fast Fibre (100Mb/s or more) by ISP
Chart 4: Average website speed for ADSL on NZ pages by Download Speed
Chart 5: Average website speed for VDSL on NZ pages by Download Speed
Chart 6: Average website speed for Fixed Wireless on NZ pages by Download Speed - indexed to the average of all three ISPs
Trend of Busy Hour Peak Speed
TrueNet runs broadband speed tests every hour of every day on internet connections. We download a file every hour of every day from all panelist connections, we have 400 panelists. Copper (ADSL and VDSL) speeds are very dependent on the length of the copper connection, ie the distance to the local exchange equipment that delivers ADSL or VDSL. ISPs cannot influence maximum speed, only line congestion, which shows as a decrease in speed on any connection at some time during the day (ToD), usually 9pm when most people are busy on the internet. We show the results at that time.
The best ISP is the one that consistently has the highest minimum hour for the month, ie they consistently provide a speed during the evening busy hour that is closest to the fastest the line will allow. On charts 1 & 2, the performance measures show the Minimum average speed of all panelists connections as a percentage of the Maximum average speed (also known as the advertised speed).
We experience many complaints from panelists of unusable performance if the average drops below about 75%, so we consider 70% the natural zero, while some users find performance is impacted if performance falls below our 90% target.
Notes that may explain the major dips in performance in previous versions:
- Netflix, the world leading internet movie supplier was introduced to New Zealand in March 2015.
- Lightbox, Spark's movie service, was introduced in July 2014
- Slingshot introduced Global Mode in June 2014
Note that for December 2016 data we have changed the measure, data past the ugly black line represents the percentage of the best speed at any time of day - ie the same as ADSL and VDSL. However we have left previous data which is based on the percentage of "Advertised" speeds, because the history has value.Chart 9: Trend of 100Mb/s Fibre & Cable Peak Speed by Time of Day, worst indexed to best speed
Rural Trend - QuarterlyChart 10: Trend of technologies tested in the Rural Sector
How these charts are built; To be fair to ISPs, we take in order;
- The median speed of tests, conducted at the same hour each day to find the representative speed of each connection at each hour for the month. Usually ~30 tests per connection
- The average of these medians for each ISP and each technology. Usually ~30 panelists
- We look for the maximum average of the medians
- We divide the average in (2) by the Maximum (3) for an index of each hour to the fastest hour
More performance comparisons here: