Bandwidth, latency and congestion

Bandwidth is a measure of the capacity of a line in kbps (kilobits per second). Latency is a measure of the speed of a line between two hosts in ms (milliseconds). Factors that affect bandwidth include the physical properties of the medium being used and whether it is shared. Cable and all wireless connections are shared media. Factors that affect latency include packet loss, your distance from DSLAMs, antennas or routers, the protocols you are using,  the medium you are using (satellite tends to have higher latency values than DSL, for example), throttling and bandwidth-shaping and packet size (MTU).

If a host uses too much of the bandwidth available to it, latency rises as a result of network congestion. A traffic analogy can help you understand network congestion. Only a limited number of cars can fit on any given road. The more cars on the road, the more slowly all them will travel. When a lot of cars arrive at a traffic light (equivalent to a router or server for the purposes of this analogy) they queue up. When the light turns green, each of them will take more time to get through the intersection than they would have on an empty road, because they are waiting for the cars in front of them to move. Congestion happens when a host sends or receives enough packets simultaneously to saturate the available bandwidth. Only a limited number of packets can fit on any given line, and when they arrive at routers or servers
they must wait in a queue to be processed. The more packets there are on the line, the longer the queue becomes, and the more slowly all of those packets will be transmitted.

Congestion-related latency may be exaggerated when an ISP responds to heavy use of its line by reducing the amount of bandwidth available to the host. Providers control access to their bandwidth with throttling and traffic shaping in order to ensure that each host on their networks only uses the explicit or implicit quota allocated to it.

TIP: Using only 90-99% of the available bandwidth will help to keep latency values low for all types of connections.

A line’s bandwidth delay product is a measure in bits (or bytes) of the maximum amount of data that can be carried on that network circuit at any given time, having been sent but not yet acknowledged. To calculate this value, multiply bandwidth (in kbps) by round-trip delay time (in seconds).

Leave a Reply