LAUSR

As one of the core directions in 6G technology evolution, computing and network convergence enables coordinated orchestration and unified management of multidimensional resources, such as computing and communication, thereby providing users with highly reliable and low-latency computing services. However, with the increasing complexity of intelligent applications, modern applications gradually exhibit structural characteristics of the coexistence of parallel tasks and multilevel dependent tasks. The intricate interdependencies among these tasks present significant challenges for the optimal allocation of computing and network resources. Therefore, this article investigates task offloading and resource allocation for applications with dependent tasks. The objective is to minimize the system’s energy consumption while meeting the application completion delay requirement. Considering the coupling between discrete and continuous variables in this NP-hard problem, we propose a joint offloading and resource allocation hybrid proximal policy optimization (JOR-HPPO) algorithm to handle the hybrid discrete-continuous action space comprising offloading decisions and resource allocation. Furthermore, we employ the Beta distribution to model the resource action space and design an action masking mechanism to optimize the resource allocation strategy in the algorithm. Simulation results demonstrate that the proposed JOR-HPPO algorithm significantly outperforms the baseline methods. Specifically, compared with the single action space optimization method, JOR-HPPO reduces the system energy consumption by $26.01~\%$ while improving the application request success rate to 30.72%.