Deep reinforcement learning based resource allocation strategy in cloud-edge computing system

• Main Authors: Ahmed Adhoni, Zameer, Habelalmateen, Mohammed I, D R, Janardhana, Abdul Sattar, Khalid Nazim, Audah, Lukman
• Format: Conference or Workshop Item
• Language: English
• Published: 2024
• Subjects: T Technology (General)
• Online Access: http://eprints.uthm.edu.my/11924/1/P17003_0e3c560e211e3d06995d79b44427688c.pdf; http://eprints.uthm.edu.my/11924/

A lot of real time processing as well as resourceintensive apps is what is needed more and thus, cloud-edge computing systems require compelling resource allocation schemes. This research focuses on the utility of Multiagent Learning framework with Deep Reinforcement Learning (MAL-DRL) which is used for solution deployment concerning resource allocation in such systems, such that the end user enjoys optimization while operators optimize resource utilization. In this work, the research focus on the simulation testing of the MAL-DRL algorithm against classical Random Allocation (RA) and singe agent DRL methods. This exhibition shows that MAL-DRL improves the average latency more (44% savings) and at the same time more resources are used (35% increase) compared both the alternatives with a combined reward measure score of 0.80. These results show that distributed decision-making and learning styles of MAL-DRL brings together faster resource allocation which can be interpreted as a reduction of users’ delay experience and therefore lead to better performance of whole system. Although the article limits simulations in areas to be covered and complexity in training, it brings the prospective benefits of MAL-DRL in the management of cloud-edge resources on the spot. In the course of this undertake, the effect of ensuring communication goals, moving learning strategies, and security measures for the ultimate goal of boosting this method's applicability in real-world antics can be explored.