Speed test graphs

A fellow IDNetter and myself are wondering if there's any extra information to be gleaned from the speed test graphs. Presumably a straight line is good, showing consistency but should multi-threaded be faster than single-threaded or is a difference expected.

For instance my latest graph with a 65Mb/s sync:

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

To me that's a near perfect connection. Client, router, modem, cabinet, ISP and server all working well. But I notice that the graphs on our new leased line at work typically have the single threaded line zig zagging either side of the multi-threaded line. Given that's fibre to our office park then fibre from a central switch to our server room and business-class Cisco kit it makes me wonder.

My fellow IDNetter is wondering if there's any significance in the fact his graph usually has the single threaded line 10Mb/s below the multi-threaded line.

Comments?

a single threaded line consistently below the multithread points to a TCP/IP stack or browser / application limitation perhaps ? or is a symptom of the congestion and "fairness" or otherwise of TCP/IP.

A BT guy wrote a paper about how "unfair" P2P apps were because their multiple TCP/IP links gave them an unfair share of bandwidth in a congested system compared to a single app download.

http://support.microsoft.com/kb/282402 says "By default, Windows Internet Explorer 7 and earlier versions limit the number of files that you can download at one time to two. Windows Internet Explorer 8 and Windows Internet Explorer 9 limit the number of files that you can download at one time to six. " but it is actually the number of concurrent connections to a server that is limited, not the "number of files" per se.

In reply to a post by Andrue:

To me that's a near perfect connection. Client, router, modem, cabinet, ISP and server all working well. But I notice that the graphs on our new leased line at work typically have the single threaded line zig zagging either side of the multi-threaded line. Given that's fibre to our office park then fibre from a central switch to our server room and business-class Cisco kit it makes me wonder.

Could be the way you speed is limited on the fttp is different from fttc eg. less or no buffering.

My fellow IDNetter is wondering if there's any significance in the fact his graph usually has the single threaded line 10Mb/s below the multi-threaded line.

Comments?

Zig zag or flat may help with diagnosis.

If it's flat perhaps it's to do with the tcp settings on the test box.

If it's zig zag it could be low level packet loss somewhere. I get lower on sinigle if I test from my desktop - but I already know my 15yo dabs value switch + home made (no crimping tool) eth connections cause 1/1000 loss (which is not enough to affect lan transfers due to the low latency, but does slow 67mbit wan).