Scheduling multiprocessor tasks in presence of correlated failures

The paper deals with scheduling multiple processor (m-p) tasks on multiple processors to maximize overall schedule reliability under the requirement that all tasks meet their deadlines. An m-p task can be represented in a schedule by one or more independently developed variants. We allow correlation of variant failures within a task and model the correlation using a hyperparameter model with Beta prior. Because the considered problem belongs to NP-hard class, four different approximation algorithms based on soft-computing methods are proposed to solve the problem. The algorithms include island-based evolutionary, neural network, population learning, and hybrid 3opt-tabu search approaches. Experiment results show that the proposed algorithms find good quality solutions in a reasonable time.