JohnButt:

 

mercutio:

 

I'd actually trust speedtest over truenet.  I suspect a lot of it is that they just have cheap bulk Vocus bandwidth, and Vocus tends to have subpar performance to a lot of destinations.

 

If they really are buying more bandwidth than other providers it may only show for parallel downloads like bittorrent, with less throughput per connection.

 

Truenet don't have enough test destinations to be unbiased in comparing ISP's depending on the paths that their upstream provider takes.

 

 

TrueNet has two (I guess that is not quite multiple) paths, AND we have our own GigE connection to both APE and WIX, the two peering exchanges in New Zealand.  Not a lot more sites worth testing in NZ though.  Speedtest on the other hand have many sites - all within the ISPs own networks, with priority traffic!  Speedtest is a test intended for testing last mile connection for fault finding - and you want that very well connected to the ISP.  But it does not necessarily traverse the ISPs network the same way ordinary traffic does which is the intention of TrueNet testing. 

 

Our testing is completely comparable between ISPs, ie we do not prioritise testing.  So if many ISPs can achieve the advertised speed at all times of day and any other ISP cannot, then it is unlikely to be TrueNet testing suite.  Our results are only published if we are confident that all, or a clear majority of panelists are seeing the same performance, in the MyRepublic case, the peak traffic period results are similar for every probe, although there is a wide range of speeds during the off-peak period, from fantastic to not so good.

 

We had similar experience with Vodafone Cable while NetFlix overloading was happening.

 

I am thinking of international destinations.  National doesn't really test out a connection well.  All things considered APE/WIX performance isn't a good test of a general users internet experience.  Things like Netflix are likely to come from ISP caches or from Sydney, and if there are significant national congestion issues then international performance can be worse affected than national.

Like back when BUBA congestion was a big issue it was generally international sites that became close to unusable during peak hours. (suttering on load, 4+ second load times etc)

There's a natural bias with TCP to prioritise low-ping destinations regardless of anything special an ISP may or may not be doing.  (so say download from a server 150 msec away at 20 megabit, download from 20 msec away destination and find that the 20 msec destination has 15 megabit/sec, and the 150 msec away destination has 5 megabit/sec)