Delay-sensitive dependency task offloading in distributed elastic optical data center networks

Abstract

Distributed optimization, learning, and inference across data centers are typical applications in the era of intelligent computing networks. These types of applications have high dependencies and are sensitive to delay. Offloading delay-sensitive dependency tasks to distributed elastic optical data center networks (EODCNs) cost-effectively is a challenging problem. To address this problem, we construct a network model, a dependency task model, and a delay calculation model. Then, to minimize the total cost of resource consumption for task offloading, we formulate an optimization problem model under various types of resource constraints. Next, we propose a delay-sensitive dependency task offloading algorithm (DSDTOA) comprising three processes: data center selection, computing and bandwidth resource allocation, and frequency slot allocation. Furthermore, we propose a theory for optimizing the split ratio of computing and transmission delays in the execution path. This can reduce the consumption of computing and bandwidth resources while meeting task delay requirements. Finally, we solve the relaxed problem model in a small-scale topology. To test the performance of the DSDTOA algorithm, we also conduct large-scale simulations under different network topologies. The simulation results show that the DSDTOA algorithm outperforms the benchmark algorithms in terms of task offloading cost and blocking probability.