Written on UA918 flying from Washington to London, and dispatched to silicon.com from my laptop using a 3G service from a car on the M25.

When we sign up with an ISP, do we really understand what broadband speeds we are going to get? Most ISPs operate a blanket get-out clause that allows them to sell broadband access with an up-to headline bit rate. You know the kind of thing: up to 8Mbps, 10Mbps, 20Mbps, 50Mbps and so on.

Judging by the amount of adverse broadband coverage and customer complaints, perhaps it is time ISPs qualified their services by at least providing a more accurate description, perhaps by adding the words, ‘sometimes’, ‘seldom’ or ‘occasionally’, to their speed figures.

Everywhere I go, I routinely check the real ISP rates by performing simple speed tests. Sometimes I am pleasantly surprised. Once or twice I have been very impressed, but mostly I’m disappointed with what is on offer.

More recently I have been conducting a series of experiments on download performance and discovered something really mean. Speed tests can be performed in seconds with consistent results at quiet times. But when I compare rates with genuine downloads taking several minutes, the results are poles apart.

In general, the broadband speed hierarchy always looks like this:

ISP advertised rate > speed test result > long-term download

In other words: long-term downloads are always significantly slower than a speed test, which in turn is slower than the advertised top rate. So what is the discrepancy? Typically, downloads are 50 per cent slower or more.

During my testing I unearthed something else, which I consider to be even sharper practice. Many ISPs appear to recognise speed-test applications and turn up the bit rate for the duration of the test. On very long downloads for big files, they progressively back off bandwidth at 50 per cent or so per notch.

What does all this look like? Here are some representative results:

speed test results graph

Results of my speed test
(Image credit: Peter Cochrane)

So a speed test may give us confidence that our ISP is delivering, but it can also reveal that through sleight of hand they are actually providing an even smaller fraction of the service they advertise, and we have to wait, and wait and wait.

Now try and imagine a world where the suppliers of automobiles, drinks, food, and clothing adopted the same strategy. Don’t worry about this actually happening, because consumer laws protect us.