Most speed tests do not report throughput. Throughput is the true performance of the user experience on the connection being tested. Throughput is the amount of data over a period of time.

Measuring efficiency of a network is important if you want to understand how the connection performs under load. As load climbs efficiency drops, just like traffic on the roads. The percentage of slowdown is essentially the drop in throughput for the load variation. Poor networks have low efficiency factors.

Max delay is the amount of time the TCP stack waits for data. Understanding when max delay is bad versus good is very important because a max delay value that is too high signifies that quality problems exist. Max Delay must not exceed the trip latency time minus the data consumption time + 20%. If max delay is very high, for example → 500 this is a good indication of timeout retransmissions. Timeout retransmission have severe penalties.

The best performing connections are one that run consistently. Meaning that the same number of bytes pass in the same interval of time every time. It does not mean it is fast or slow. 1 byte per minute would be consistent but also very slow. Inconsistent data flow is a sign of data flow issues such as packet loss, duplicates and retransmissions. It can also mean that the connection is provisioned poorly. Example 20Mbps could be delivered as 100Mbps for 1 second of time and then nothing for 4 seconds of time. The average being 20Mbps. Not very good for media based applications.

TCP forced idle defines the amount of spare capacity that is available for a specific connection under test which results from the capacity limit being greater than the amount of the TCP unacknowledged payload (called a window). This is the same as an idle road that has little traffic. Its not roads fault that only a few cars are using it.

Duplicates and part duplicates are bad problems and imply that retransmissions are occurring. Duplicates occur when packets are delayed but never get lost resulting in a retransmission and both packets eventually arise. Whichever packet arrives last (although it can be the first packet sent) will be flagged as duplicate. Part duplicates are more problematic than non-part duplicates because it implies that packets are being fragmented.

There are two types of retransmit… A fast retransmit and a timeout retransmit both are problems that cause slow connections although the retransmit timeout is the worse. The best way to describe a retransmit is to consider two people talking on the phone are repeatedly ask the other to repeat what they just said because they couldn’t hear it. Makes it hard to have a good conversation experience.

Out of order packets are only measured in a quality test. An out of order packet is a problem because it slows the data flow and can cause network timeouts and duplicates. These are significant problems if they occur.

Bytes that arrive outside the TCP leading or trailing edge will be discarded as outside window. This is a serious problem because it means the data has no place to go. A bit like driving into a multi-story car park which broadcasts at the entrance that there are spaces available when there aren’t any.