Joint optimization of resource allocation and tasks scheduling in RDMA-enabled edge computing networks

Abstract Remote direct memory access (RDMA) enables ultra-low-latency and energy-efficient data transmission in edge computing networks, thereby facilitating coordinated task scheduling and resource allocation. This paper investigates an RDMA-enabled edge computing system with multiple edge servers coordinated by a central scheduler, where computation tasks arrive stochastically and are dynamically assigned. The objective is to minimize the long-term energy consumption while ensuring queue stability in the presence of heterogeneous task characteristics and varying system conditions. To tackle this long-term stochastic optimization problem with coupled task scheduling and resource allocation, a Lyapunov optimization framework is adopted to transform it into a sequence of per-slot deterministic subproblems. A closed-form optimal solution is derived for computation frequency, and a low-complexity water-level balancing algorithm for task scheduling is proposed based on convex relaxation and Karush–Kuhn–Tucker (KKT) analysis. The resulting online algorithm operates without requiring prior knowledge of task arrival statistics. Simulation results demonstrate that the proposed algorithm effectively reduces long-term energy consumption while maintaining queue stability.