Parallel Efficiency

Parallel Efficiency measures the overall efficiency of splitting computation over processes and threads.

Parallel Efficiency is the ratio of Avg(Useful) to Total Runtime. For ideal parallelisation, i.e. perfect load balance and no overheads introduced by the parallel libraries, the ideal runtime would be Avg(Useful).

In the additive hybrid approach the sub-metrics of Parallel Efficiency (PE) consider efficiency of processes (e.g. MPI) and threads (e.g. OpenMP) separately, i.e. Parallel Efficiency splits into Process Efficiency and Thread Efficiency.

\[Parallel Efficiency = \frac{Avg(Useful)}{TotalRuntime} = Process Efficiency + Thread Efficiency - 1\]

In order to fully understand the formulas, you may also visit the glossary of the metrics terms.