Wireless is becoming less of an issue with wireless AC. I certainly see people getting 200Mbps over wireless on virginmedia regularly on iPads etc.

In reply to a post by MrSaffron:

We see people nailing the 76 Mbps so have never seen the point of making a big noise over the percentage
http://www.thinkbroadband.com/speedtest/results.html...

Maybe it's a linux/seamonkey thing, but I've always got saner results from your flash test (though there is one issue with it).

I guess testing data throughput with tcp is always tricky, but the new test is clearly capable of giving impossible results - just look at this freak, both avg and burst upload being impossible for a 20 meg sync ptm.

http://www.thinkbroadband.com/speedtest/results.html...

On the 76 thing, I have no idea what it is, but with adsl bt iprates were actually defined by bt as aal5 payload (so ip +2 ppp). On adsl the (so called) ip profile was even less related to ip, though IIRC bt docs did call that one bras rate = atm payload which did historically reduce speed below syc rate. Not sure about 21cn % rates - also not sure whether on fttc the profile is slightly limiting (being on plusnet with a locked down ECI modem it's hard to test that).

If the profile doesn't limit then on 80 sync it seems to calculate (accepting my sums may be wrong) to ip payload. Of course if bras limits then it it could be something else.

If talk talk test better maybe it means BT do limit - but then do TT use pppoe? I don't know but if they don't that could account for it. I came across a mail searching once that said sky don't use pppoe and it also said something about them not wanting to send fcs - if LLU providers avoid that + pppoe that's 12 bytes per packet saved which will easily show @ 80meg.

Upload yes, and its pretty clear from the graph that upload is spiking up/down, if I added smoothing issue would go away, but prefer to avoid things like smoothing or interpolation as these can hide useful bits of data.

Fair enough, though given the speed I bet the upload leaving my line was rock solid max rate at ip level, with the bursting being a tcp artifact/feature due to "me" dropping packets off the end of a buffer somewhere.

Out of all speed tests for upload I really rate your flash test, it pretty much gives consistent results, is single threaded and so lets me "play" like turning off wscale. It seems to be quite accurate eg. flipping timestamps on/off the difference seems correct from what I expect.

Saying that I may just be biased because it agrees with my calculated rates - it may be totally wrong and just re-enforcing my misconceptions

The flash had the smoothing turned on some years ago which explains a lot.