Using A3C learning and residual recurrent neural networks, dynamic scheduling for stochastic edge-cloud computing environments

PROJECT TITLE :

Dynamic Scheduling for Stochastic Edge-Cloud Computing Environments using A3C learning and Residual Recurrent Neural Networks

ABSTRACT:

The widespread adoption of applications that are based on the Internet of Things (IoT) has led to the emergence of the fog computing paradigm, which makes it possible to seamlessly harness resources from both mobile-edge devices and the cloud. Constrained resource capabilities, mobility factors in IoT, resource heterogeneity, network hierarchy, and stochastic behaviors all make it difficult to schedule application tasks efficiently in these kinds of environments. The currently available heuristics and reinforcement learning-based approaches lack the ability to generalize their findings and quickly adapt to new circumstances, which prevents them from providing an optimal solution to this problem. They are also incapable of utilizing the temporal workload patterns and are only suitable for centralized setups due to this limitation. Asynchronous advantage actor critic, or A3C, learning, on the other hand, is known for its ability to rapidly adapt to dynamic scenarios with a smaller amount of data, and residual recurrent neural network, or R2N2, is known for its ability to rapidly update model parameters. As a result, we propose an A3C-based real-time scheduler for stochastic Edge-Cloud environments. This scheduler will make it possible for multiple agents to engage in decentralized learning simultaneously. In order to make effective decisions regarding scheduling, we make use of the R2N2 architecture, which allows us to record a significant number of host and task parameters along with temporal patterns. The model that is being proposed is adaptive and has the capability to tune various hyper-parameters based on the requirements of the application. Through the use of sensitivity analysis, we explain why we chose the hyper-parameters that we did. When compared to the state-of-the-art algorithms, the experiments that were carried out on real-world data sets show a significant improvement in terms of energy consumption, response time, Service-Level-Agreement, and running cost by 14.4, 7.74, 31.9, and 4.64 percent, respectively.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here