| Summary: | 碩士 === 國立交通大學 === 電機工程學系 === 107 === Fog computing is the technologies which are proposed to overcome the long latency of cloud computing. By deploying fog nodes near users, fog computing is able to reduce the latency and improves quality of service level. However, due to the limited computing resource of single fog node, the resource allocation algorithm for fog computing is crucial to the system performance. Linear programming(LP) is a mathematical tool for modeling optimization problem like resource allocation. The optimal solution can be found by defining objective function and constraints that consist of inequalities or equalities, but the high complexity makes the execution time of LP model the bottleneck of the system.
This thesis proposes a near-optimal resource allocation framework for fog computing (AR/VR recognition) by using deep reinforcement learning. Experimental results show that the proposed mechanism outperforms existing methods in terms of balance effect, on-line computation latency, and scalability.
|
|---|
