- Aug 23, 2013
Your argument is what exactly though? My ISP is responsible for that peering as well, I used to get my 90Mbps to Europe on Vox + OpenServe, on CI I can only get over 30Mbps consistently if I use the proxy, meanwhile I get barely 15Mbps:TCP/IP was invented in the early seventies for initial use on a wireless network. It is the same thing we use today and an analogy would be that we are using a method that works for throwing bean bags but we expect the same reliability using paint ball guns. It is well past its sell by date.
Thus the stack on your PC, router, interconnecting devices and destination host all play a part. As an example, if you are on Windows 7, use a cheap router with realtek chips then don't be alarmed at poor speeds.
So besides the fibre operator's network which itself might have an access, distribution and core with different capacities, the ISP network would also be dependent on cross connects to peering partners. There are more than a dozen points of congestion along the path which could include dodgy fibre, dodgy chips, as well as misconfiguration and even buffer bloat. All these factors contribute to it be known as broadband and best effort.
I'll stick out my head on the irrationality of the logic being presented. Most people are prepared to accept the evidence of their cr@ppy speedtest conducted off the end of a cr@ppy broadband connection as the ISP being a problem but won't accept a speedtest on the ISP core network which shows no symptoms of problems? Why?
And my upload is way higher, and that should be suffering due to poor scaling as well.
I know how TCP/IP scaling works, and it has a huge impact on single-threaded, but on multi-threaded, that window of time changes substantially since we're talking 3-8 threads scaling, with TCP/IP starting its scaling at about 5Mbps per thread if over 90ms, with it definitely increasing past that point if going higher, so I should see a minimum starting scaling of ~15Mbps and I should definitely be able to max my line with 8 threads.