Efficient stochastic scheduling for highly complex resource placement in edge clouds

For the edge cloud-based large-scale distributed systems in wide areas, it is important to quickly adjust the deployed resource in multiple edge clouds to maximize the resource revenue and meet the quality of service requirements. The mean values of demands are usually used in the scheduling algorithms for simplicity, but in the real-world scenarios the resource demands may fluctuate greatly, which cannot be effectively modeled in the mean value-based demand model, resulting in the under-utilization of resources. To address the problem, we investigate a general stochastic scheduling problem in the edge clouds, whose objective is to place the given amount of resources into the edge areas, and to maximize the scheduling revenue like the weighted sum of satisfied demands. We then propose an efficient algorithm by identifying the optimal conditions of nested subproblems. Experiments show that in the scenarios with general settings, the algorithm can achieve at least 97% average revenue of the traditional optimal algorithm with much lower time complexity, which can be further reduced through parallelization. The algorithm has the potential to be an effective supplement to the existing algorithms under the time-tense scheduling scenarios with a large number of resources.