Learning Based Genetic Algorithm for Task Graph Scheduling

Algorithm 1

A new task graph scheduling algorithm.

Step 1:
- Create the initial population (classic chromosomes) as described in Section 3.1.
Step 2:
- While termination condition has not been satisfied Do
- For each classic chromosome in the current population Do
- Convert it to the extended structure chromosome (such as Table 2).
- Enhance chromosome by applying learning operators (reward and penalty) on the chromosome until the
makespan could be less, described in Section 3.4.
- Convert extended chromosome to classic chromosome by removing rows related depth and probability
- Apply the standard roulette wheel selection strategy for selecting potentially useful individuals for recombination
- Apply the standard two-point crossover operator on processors in two enhanced chromosomes
- Apply the standard mutation operator on processors in two enhanced chromosomes
- Apply the reuse idle time heuristic on 30% of best chromosomes, described in Section 3.3.