At the start of July we looked at what evidence there was for congestion on the largest providers using our main speed test and now we have analysed the data from a more sensitive version of the tester.
The speed test these results are based on is our flash based speed test which is available to registered users on www.thinkbroadband.com/speedtest or can be accessed directly by its alternate URL if you are not logged in on thinkbroadband. With the number of people turning off flash plug-ins we have also released a version of our mobile and tablet compatible speed test that repeats the single thread test.
So what is the difference between a single thread and a multiple thread speed test? The multiple thread test should ramp up to your lines maximum speed much faster if you have a poorly configured RWIN value and should be better at saturating a connection in cases where there is some other activity. This all means that when you do a single thread download (more akin to saving a file from a website) the test is more sensitive to both your devices configuration but also any congested links on the journey from your connection to our server.
The previous analysis for 5 of the 6 largest providers showed roughly what you would expect with a reduction in speeds at peak time, but PlusNet bucked the trend which was unexpected and also goes against some very long discussions amongst users about peak time speed issues, hence a further analysis using our more sensitive test.
As the sample volume on the flash tester was smaller we have dropped EE from the analysis, the reason being that while we have plenty of anonymous user data the volume using the full flash test was too small over the last few months to read anything into the figures.
|Table of Average Median Download Speeds for Providers in June 2015|
7am to 3pm
3pm to 6pm
6pm to midnight
|BT||16.6 Mbps||19 Mbps||16.9 Mbps||+1.8%|
|PlusNet||12.7 Mbps||16.4 Mbps||21.3 Mbps||+67.7%|
|Sky||11.8 Mbps||12.3 Mbps||11.7 Mbps||-0.9%|
|TalkTalk||10.8 Mbps||9 Mbps||8.6 Mbps||-20.4%|
|Virgin Media||45.2 Mbps||39.1 Mbps||31.9 Mbps||-29.5%|
|Table of Average Mean Download Speeds for Providers in June 2015|
7am to 3pm
3pm to 6pm
6pm to midnight
|BT||23.8 Mbps||23.3 Mbps||24 Mbps||+0.9%|
|PlusNet||20.5 Mbps||23.5 Mbps||26.3 Mbps||+22.1%|
|Sky||17.8 Mbps||17.3 Mbps||14.1 Mbps||-26.2%|
|TalkTalk||15.6 Mbps||14.7 Mbps||14.9 Mbps||-4.7%|
|Virgin Media||53.6 Mbps||48 Mbps||41.7 Mbps||-28.5%|
And we end up with a conundrum again where PlusNet seem to be able to increase speeds overall during peak time, but we know that some individuals (guesses of speed issues affecting 1 in 10 on Plusnet are doing the rounds, and many find reconnecting that puts them on a different gateway solves the problem) are seeing the opposite. Exactly why this is the case is difficult to say, maybe the daytime users on PlusNet are those on longer ADSL2+ or VDSL2 lines, or PlusNet traffic management switches to a different mode at some point in the afternoon helping tests to perform better, but then not everyone sees that effect. One theory doing the rounds and this is supported by a limited set of data from other BT Wholesale based providers is that the BT Wholesale network is throwing up some oddities, due to capacity issues or how the traffic is handled. The reason this does not affect BT Retail connections so heavily is they utilise a slightly different set of products from BT Wholesale – other providers are welcome under the equivalence rules to use that product range, but for various reasons it only appeals to BT Retail.
The dips for Sky, TalkTalk and Virgin Media appear much what we would expect and in fact the Virgin Media data will probably have a lot of users nodding that they see speeds drop at peak times, of course when you are selling 50 Mbps to 152 Mbps products you can afford a big drop in throughput so long as packet loss is kept to a minimum.
The swings on the PlusNet graph seem to vary a lot, which does support the idea that maybe we are seeing a very different cohort of PlusNet users, though do know and can see some users who come home and find speeds are slow jump a gateway and then get good speeds for the rest of the evening, which could just as easily explain the variation. We could potentially test a cohort of PlusNet users with a test that automatically repeats itself in a browser window every 30 minutes for example, and while a browser cannot tell whether the connection is busy we can monitor the buffer bloat to infer if another device is causing the connection to slow down.
The aggregate plot above (not using any of the providers mentioned in this article) shows what is possible when you have enough test data from a single connection and it is clear to see that the average download speed lies in the 17 to 20 Mbps region. By plotting the throughput during the time of the test you can also get a good understanding as to things vary over time. When we revamp the speed test result handling this sort of plot is one option we are considering for registered users. A few words on the green plot, the cluster of tests around 24 to 25 Mbps reflects the time of year when the FTTC DLM system seems to relent and allow the line to go faster.
The TalkTalk chart is interesting as it seems to show that the afternoon and evenings are more consistently slower than the main daytime window and reflects the general understanding for how broadband usually performs across the day.
The Virgin Media graph shows pretty much what we bet everyone was expecting from all the providers and its consistency over time seems to suggest that we have not got a broken analysis method, but that external factors to our speed test are behind the variability for the other major providers. One observation on the Virgin Media results is that as the average speeds are increasing the size of the dip at peak times is also increasing which is a combination of the realities of the shared coax local loop on DOCSIS technology and the global reality that as you push for higher headline speeds the peak time dip can be larger – when you are a user on an old slow 3 Mbps a dip down to 2 Mbps is a real annoyance as streaming will buffer horribly, but if on a 100 Mbps connection a dip to 60 Mbps should still result in good video streaming.