Gustafson's law

E783773

Gustafson's law is a principle in parallel computing that argues overall speedup can scale with the number of processors by increasing problem size, challenging the fixed-workload limitation implied by Amdahl's law.

Try in SPARQL Jump to: Statements Referenced by

Statements (40)

Predicate Object
instanceOf performance model
principle in parallel computing
scaling law
addresses scalability of parallel programs
speedup in parallel computing
alsoKnownAs Gustafson–Barsis law NERFINISHED
applicableTo distributed-memory systems
shared-memory multiprocessors
argues overall speedup can scale with number of processors
assumes fixed execution time
increasing workload with processor count
problem size grows with processor count
scaled problem size
category laws of computer science
parallel computing theory
challenges fixed-workload limitation of Amdahl's law
contrastsWith Amdahl's law NERFINISHED
defines N as number of processors
S(N) as scaled speedup
α as serial fraction of execution time
emphasizes practical workloads rather than fixed-size benchmarks
field high-performance computing
parallel computing
focusesOn weak scaling
implies large problems can use many processors efficiently
parallel fraction can dominate as processor count increases
influenced design of large-scale scientific simulations
thinking about massively parallel processing
motivatedBy limitations of Amdahl's law for large-scale systems
proposedBy John L. Gustafson NERFINISHED
publicationYear 1988
publishedIn Communications of the ACM NERFINISHED
relatedTo Amdahl's law NERFINISHED
parallel speedup
strong scaling
weak scaling efficiency
speedupFormula S(N) = N − α (N − 1)
usedFor designing scalable parallel algorithms
estimating parallel performance
evaluating supercomputer scalability

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

Amdahl's law contrastedWith Gustafson's law