The bandwidth of the medium limits the speed at which you can communicate. The greater the bandwidth the greater the data rate.
Units of bandwidth are hertz, or 1/s.
Since bandwidth (and some details like coding schemes, and channel noise) determines the raw data rate possible, some authors use bandwidth to mean the raw data rate of the channel. In this case the units are bits per second.
The bandwidth of the channel, from this perspective, is the raw rate at which you can signal data into the channel. It does not take into account the need to get acknowledgments or other overheads.
Units of throughput or data rate are bits per second.
Throughput depends on bandwidth (fundamentally), and efficiency of communication.
The efficiency of the communication depends on how the protocols are designed, and how well the communicating parties make use of the raw bandwidth of the channel.
There are four types of delay experienced by a packet on its way through a packet-switched network.
Often times it is twice the channel length that matters to us (as in sending a packet and waiting for an acknowledgment - two round trips). This is known as RTT.
Whether you care more about throughput or latency depends on the application.
An interesting measure of which performance term is significant is the number of instructions that could be computed in the time spent sending a message. Very fast computers means being able to execute many instructions in the time spent waiting for a response from a message. This suggests that the computation/communication ratio be skewed towards doing more computation.
This delay depends on the amount of traffic arriving at the node, and the rate at which traffic can leave the node.
Average arrival rate: a packets/sec
Length of a packet: L bits/packetSo the rate of arriving bits is L * a.
If the rate at which bits leave (the throughput out of the node) is R, then we better make sure that L * a is less than R or the queue blows up.
The magic ratio of L * a / R is known as the traffic intensity.
As traffic intensity approaches 1, the queueing delay knees up exponentially.
Geometrically this can be envisioned as a pipe (cylinder). The length of the pipe is proportional to the latency or delay in the channel. The width or diameter of the pipe is proportional to the bandwidth or througput of the channel.
The actual product (when the units are seconds and bits/second, respectively) is the number of bits that will fit in the channel.
The delay bandwidth product of the channel is interesting because it indicates the storage capacity of the channel. If you start sending data, then the capacity of the channnel tells you how many bits will be sent before the first bit even reaches the receiver.
Another view is that if you are not able to send an amount of data equal to the delay x bandwidth product, and you must then stop and wait for an acknowledgment, then you have a measure of channel utilization. Tiny amounts of data over a long or fast channel means bad utilization.
How long is a bit?
Each bit occupies the sender for (1 / data rate) seconds to signal it. Signalling at 1 bps means each bit takes 1 second to signal. Signalling at 1000 bps means each bit takes 1/1000 s or 1 ms to signal.How far does the leading edge of the bit travel in the time it takes to signal the entire bit? That depends solely on the veolcity of propogation, not the the signalling rate. The two together determine how "long" a bit is.
If the velocity of propogation is 1 m/s, then the leading edge of the bit is 1 second ahead of the trailing edge for the 1 bps data rate, then a bit is 1 m long. If the signalling rate is 1000 bps, then the leading edge has only travelled 1ms, and the entire bit is 1 ms * 1 m/s = 1mm long.
Throughput (versus bandwidth) depends on how efficiently we communicate, and on the RTT.
Putting these together we have