I decided that I ought to just get a big file from somewhere and download it and time it.
What's the easiest way to do it on an iPad with a straight TCP-based download? Something in a Web browser presumably.
I used a 50 'MB' (whatever that means)
download test file on the thinkbroadband website and downloaded that on my iPad in Safari and that took ~57s (with an error bound of 2 secs diameter) so I make that 7.02 Mbps downstream if we assume that they mean ‘MB’ = 10
6 = 1000000 bytes = 1 true MB in the correct notation as they didn't say MiB for 2
20.
I did a number of downloads, with 50MB and 100MB sizes and took the fastest result, as any slow-downs have to be deficiencies in the test set up, congestion, delays in the server, other traffic and so-on. I tried to get an estimate of the error due to TCP start time but failed due to variation in the results. I removed one-off delay due to DNS lookup by repeating tests so ensuring that DNS queries are cached.
This is to be compared with a downstream sum-of-IP-TX-rates of 7552566 bps reported by AA the ISP summed over the three lines and the far less meaningful 8591kbps is the sum-of-downstream sync rates. So that means TCP payload rate is 93% of the AA TX rate and since IPv6+TCP headers cost 4% then that isn't bad, TCP being better than 97% efficient compared to 100%=the maximum payload rate TCP could possibly achieve, calculated by knocking out TCP+IPv6 headers. It could have been IPv4, I didn't check, mea culpa. The truth will be a bit better than 93% because slow start was not removed from this.
It also means that (triple) multi line IP-bonding works fairly well, costing rather less than 4%, but otherwise quite unknown from these numbers.