GPU-based Parallel Implementation of Swarm Intelligence Algorithms

Chapter 4

Performance Metrics

Contents

4.1 Parallel Performance Metrics  49

4.2 Algorithm Performance Metrics  50

4.3 Rectified Efficiency  51

4.4 Case Study  51

4.4.1 Case Study on Parallel Performance   52

4.4.2 Case Study on Algorithm Performance   53

4.5 Summary  54

4.1 Parallel Performance Metrics

Speedup and efficiency are the most common metrics used by the parallel community. Speedup can be defined as the ratio of the execution time of parallel implementation (T_p) and the execution time of the sequential one (T_s) (Eq. (4.1)), and efficiency is the normalized value of the speedup regarding the number of cores (m) executing a parallel algorithm (Eq. (4.2))

$S=\frac{T_s}{T_p}$

  (4.1)

$E=\frac{S}{m}$

  (4.2)

In conventional parallel computing, the speedup allows ...