With the widespread application of distributed computing in the field of mobile applications, mobile devices, due to their limited computing power, storage, and energy, find it difficult to efficiently handle complex application tasks. Moreover, the massive heterogeneous tasks in the distributed environment also pose challenges for resource management. This paper proposes a distributed resource management mechanism based on deep reinforcement learning, aiming to address the resource allocation and task processing issues of mobile applications in a distributed environment. Through fine - grained application partitioning, mobile applications are accurately divided into multiple subtasks, and computationally intensive subtasks are selectively offloaded to reduce the burden on local devices. Meanwhile, a multi - level deep reinforcement learning model is constructed. This model dynamically adjusts resource management strategies according to task characteristics and optimizes the partitioning strategy through model feedback, forming a closed - loop optimization system. Experiments demonstrate that this mechanism significantly outperforms other methods in key indicators such as resource utilization, data transmission delay, and task completion success rate.