Network Metrics

From 118wiki

Jump to: navigation, search


Network Metrics

An important topic in any study of communication networks is how to measure and analyze properties of architectures, protocols, software, and hardware. This page scratches the surface of this topic, introducing some basic terminology.


The term bandwidth is very widely used, and abused. Originally, this referred to a range of frequencies, that is, a "band" of the radio frequency spectrum. Now we usually think of bandwidth as the amount of data per unit of time transmitted over a channel, link, connection (or whatever term is meaningful for the layer of abstraction we focus on). At the physical layer, bandwidth may only mean transmission rate, however at the physical layer there can be a difference between bits and signals or symbols (for instance, baud does not mean "bit").

Further, bandwidth available at the physical layer may not be what we observe at the application layer, due to various overhead factors. So we can refer to "raw bandwidth" as provided at the physical layer versus "effective bandwidth", which is what we observe at the higher layers. The term throughput is possibly a better term than bandwidth for measuring data rate. Even throughput numbers can be deceptive because, in practice, data could get corrupted and have to be retransmitted: the term goodput has been suggested for what we observe after taking retransmission into account.

The Numbers. What are the units for measuring bandwidth, throughput, etc? There are some traditions here:

  • Transmission rates, throughput, and bandwidth are measured by bit rate. Note carefully that powers of ten are used for prefixes in the terminology.
  • Data (measuring file size, memory size, disk capacity) is measured in bytes. Note carefully that binary powers of 10 are used for prefixes in the terminology.


After transmitting some number of bits, there can be some delay before the transmitted bits are received. Latency is the catch-all term for the total delay, from initiation to termination, of some network operation. By "network operation" we can mean the sending and receiving of one packet, the request generation and eventual output display of a web page, or the time to start and finish a download of a video. Latency can be measured either one-way or two-way. Latency is measured in seconds (or minutes, hours, if we are not so lucky).

What are the components of latency? One is the time taken for transmission (see above; it depends on how much data is involved as well as bandwidth). Others include propagation delay and queueing delay (queueing delay is a catch-all for the time taken by all the layers, which add processing overhead for protocols). While we may be able to precisely calculate transmission delay, the numbers for propagation and queueing delays are likely to be approximate, based on expected values observed from experiments.


Though not really a term for network measurement, many examples in network channels show asymmetric properties, which have to be known before we can analyze a given situation. The term simplex communication refers to links that can only be used in one direction (typically, fiber has this property). Communication can be half-duplex or full-duplex on a link. Of course we can always "create" a full-duplex link by putting two simplex links (in opposite directions) together.

Another example of asymmetry is ADSL, which is a telephone protocol dividing up bandwidth unequally between the two directions. Many broadband protocols also split up bandwidth unequally to suit their intended applications.

Other Metrics

There are many other network metrics (some mentioned in metrics (networking)) that we leave for later in the course. This include number of hops in a network path, hit rates for web caches, and other resource statistics.

Personal tools