This invention discloses a latency-constrained scheduling method for heterogeneous computing resources in a computing power network, belonging to the field of communications. It includes the following steps: obtaining the number of computing tasks, data volume, computational load, and latency constraints; obtaining the performance parameters of the heterogeneous computing servers; and obtaining the
transmission bandwidth and transmission power of the
communication channel. Following the basic idea of jointly scheduling and optimizing
total energy consumption for communication and computation, the scheduling strategy and
server performance parameters are iteratively optimized alternately: on the one hand, the scheduling strategy is fixed, and the
server performance parameter configuration is optimized using dynamic
voltage-
frequency scaling technology; on the other hand, the
server performance parameters are fixed, and the scheduling strategy is updated using the projection
gradient method. The scheduling method can provide an energy-optimized heterogeneous computing
resource scheduling strategy and server performance parameter configuration. This invention can be used for the scheduling optimization of heterogeneous computing resources in a computing power network to achieve the beneficial effect of reducing
energy consumption for completing computing tasks under latency constraints.