CODES OF PAPER: E2E Resilient and Proactive Resource Management with Network Slicing

Intelligence and flexibility are the two main requirements for next-generation networks that can be implemented in network slicing (NetS) technology.This intelligence and flexibility can have different indicators in networks, such as proactivity and resilience. In this paper, we propose a novel proactive end-to-end (E2E) resource management in a packet-based model, supporting NetS. Since guaranteeing quality of service (QoS) in NetS has many challenges, we present an intelligent method that has two characteristics: resilience and proactivity. Guaranteeing successful slice provision is costly, we formulate a comprehensive model of the imposed costs. To minimize the cost function, we introduce a new optimization problem with radio, processing, and transmission resource constraints. In addition, we introduce two new constraints that guarantee the proactivity and resilience capabilities of the network based on the probability of successful slice provisioning (PSSP). Since the proposed optimization problem is non-convex, online and belongs to the NP-hard category, we adopt a deep reinforcement learning (DRL) based method to solve it. The obtained results reveal that the applied method can improve the percentage of successful slice provisioned (PrSSP). In addition, the resiliency time is reduced comparatively. Finally, as the main achievement, the resilient scenario improves PrSSP compared to the non-resilient scenario.

智能与灵活性是下一代网络在实现网络切片（NetS）技术中的两大核心需求。此种智能与灵活性在网络中可表现为多种指标，诸如积极性与弹性。本文提出一种新颖的基于数据包模型的主动式端到端（E2E）资源管理方案，以支持网络切片技术。鉴于在NetS中保证服务质量（QoS）面临着诸多挑战，我们提出了一种兼具弹性和积极性的智能方法。确保切片成功分配的成本高昂，因此，我们构建了一个全面的成本模型。为了最小化成本函数，我们引入了一个新的优化问题，该问题包含无线、处理和传输资源约束。此外，我们还引入了两个新约束，以确保基于成功切片分配概率（PSSP）的网络具有积极性和弹性能力。由于所提出的优化问题是非凸的、在线的，且属于NP难类别，我们采用基于深度强化学习（DRL）的方法来解决它。所得结果揭示，所采用的方法能够提高成功切片分配的百分比（PrSSP）。此外，相较于非弹性场景，弹性场景显著降低了恢复时间。最终，作为主要成果，弹性场景相较于非弹性场景在PrSSP方面有所提升。