Amdahl's law

Amdahl's law gives the maximum speedup from improving part of a system when another part remains unchanged. For parallel computing, it limits speedup according to the fraction of a program that must run serially.

Speedup: The ratio of the time required for a task before and after a change. Amdahl's law calculates the maximum speedup an optimization can deliver.
Parallelism: The ability to perform multiple operations at the same time. Amdahl's law measures how much parallelism can accelerate a fixed workload.
Gene Amdahl: An American computer architect and engineer known for designing mainframe computers and formulating Amdahl's law. He articulated the scaling argument that bears his name.
Performance engineering: The practice of measuring and improving a system's speed, efficiency, and resource use. Amdahl's law helps prioritize improvements by estimating their maximum payoff.
Universal scalability law: A model of system scalability that includes contention and coordination costs as resources increase. It models scaling penalties that Amdahl's simple formula leaves out.
Serial fraction: The portion of a computation that cannot be divided among parallel workers. This fraction sets the asymptotic ceiling on parallel speedup.
Fraction: A part expressed as a proportion of a whole. The law separates total work into serial and improvable fractions.
AFIPS Spring Joint Computer Conference: A 1967 computing conference where Gene Amdahl presented an argument about parallel processing limits. Amdahl introduced the law in a paper presented at this conference.
Multicore processor: A processor containing multiple independent processing cores on one integrated circuit. Its cores can accelerate only the program work that runs in parallel.
Bottleneck: A system component whose limited capacity constrains the performance of the whole system. The serial fraction acts as a persistent bottleneck in Amdahl's model.

Linked from 12 pages