HomePage | Recent Changes | Preferences

The engineering definition of latency is how long a given message takes to traverse a system. For example, the latency of a local telephone call is several milliseconds: it takes the message (my voice) several milliseconds to traverse the telephone line to the recipient. The latency of an ordinary letter sent overseas could be measured in days or weeks.

Latency is often used in computer systems along with the concept of throughput. A simple model may assume that throughput equals 1/latency. For instance, if it takes my underwear 2 hours to wash and dry, I could certainly achieve a throughput of 1/2 (1 load per 2 hours or 1/2 load per hour).

Engineers use the concept of pipelining to increase throughput for a given latency. Assume it takes 1 hour to wash and 1 hour to dry. When my underwear starts drying I may start washing my shirts. When the shirts start drying I can start washing my pants. It is still true that the latency of the system is 2 hours (any given item will take 2 hours to traverse the system). However, the throughput is now up to 1 load per hour (every hour produces another finished load of clothes).

The concepts of latency and throughput have a dramatic impact on the design of communications and computing systems.

see also rotational delay, which is another form of latency

HomePage | Recent Changes | Preferences
This page is read-only | View other revisions
Last edited October 14, 2001 1:47 am by 24.49.40.xxx (diff)