When it comes to broadband, the “speed” of your line is not always what’s written on the box.
We usually use speed to refer to throughput, or how fast you can download or upload data over a connection. However, there is an important measurement of a line’s performance which is not always highlighted – your latency, also known as ping.
Latency refers to the time it takes for a data packet to arrive after it has been dispatched.
For example: It may take a packet 4 milliseconds to travel from a server in Johannesburg to your device if you are in Gauteng. If you’re in Cape Town, it could take that packet 15 or 20 milliseconds to reach you.
When you access servers overseas, these latencies rise sharply. To a server in London, you can expect latencies of at least 150 milliseconds, while a server in New York will see latencies of at least 250 milliseconds and up from South Africa.
With throughput – the amount of data your line can carry – you can always upgrade to a “faster” line to increase its capacity. Moving from a 10Mbps line to a 100Mbps line, for example.
There is usually very little you can do about latency, however, as it is physically limited by the speed of light in the case of fibre.
This does not mean latency and throughput are not related, though.
While they may appear to behave independently of each other, when you start looking at 400Mbps and faster fibre connections, their effect on one another becomes apparent.
The limits of TCP
Transmission Control Protocol (TCP) is part of the foundation of the Internet. It helps ensure that data ends up where it must go.
It does this through a system of sending and waiting for data to be acknowledged. Until you acknowledge the sender, it won’t send the next batch of data through.
After enough time has passed without an acknowledgement, it is assumed the data was lost in transit, and it is retransmitted. This is called packet loss.
Since there is a maximum size for the amount of data that can be sent over TCP before waiting for an acknowledgement (the TCP Receive Window), the amount of time you have to wait for that acknowledgement can impair throughput.
The maximum transmission unit over TCP is typically 1,500 bytes. Including the TCP/IP header information then leaves you with a maximum segment size of usually 1,460 bytes.
These values vary from network to network, but using the measurement above gives us an ideal scenario.
Calculating the maximum theoretical throughput of your line also uses the round trip time, which for the purposes of this article will be double the expected latency.
We used the TCP throughput calculator on SWITCH.ch to calculate the theoretical maximum throughput of a fibre line in Johannesburg to local and international servers based on latency – as detailed below.
It assumes packet loss of 0.000001%.
As the results show, even with a 1Gbps line, a user will not get download speeds of over 500Mbps from most US-based servers.
|Maximum download speeds over a fibre line in Johannesburg|
|Server Location||Latency (Ideal)||Throughput*|
|* The maximum theoretical throughput over a line.|