In this paper, we propose strategies and develop solutions for a network service provider (NSP) to cost-effectively provision and manage a large number of network slices. Specifically, we propose a novel framework, namely, network slice bundling, in which (1) an NSP can allocate resources and create multiple network slice bundles in advance, (2) a network slice request can be quickly instantiated in the bundle that supports its service requirements, and (3) network slices in the same bundle can share the resources and achieve a multiplexing gain by leveraging the stochastic behaviors of resource usage. Within this framework, we focus on a core problem, which is how to leverage the multiplexing gain to maximize the utility by optimally assigning multiple network slices to a set of pre-defined bundles. We formulate an optimization problem and theoretically analyze the irregularity of constraints and the difficulty of the problem. We develop a novel reinforcement learning (RL) based slice assignment solution. Finally, we conduct extensive data-driven simulation experiments. The numerical results confirm that the proposed solution can efficiently solve the network slice assignment problem and achieve significantly higher utility than the best baseline algorithm.

中文翻译：

---

利用网络切片捆绑中的复用增益

在本文中，我们为网络服务提供商（NSP）提出了策略并开发了解决方案，以经济高效地提供和管理大量网络切片。具体来说，我们提出了一种新颖的框架，即网络切片捆绑，其中（1）NSP可以分配资源并预先创建多个网络切片捆绑，（2）网络切片请求可以在支持其的捆绑中快速实例化服务需求，以及（3）同一捆绑包中的网络切片可以共享资源，并通过利用资源使用的随机行为来实现多路复用增益。在此框架内，我们关注一个核心问题，即如何通过将多个网络切片最佳地分配给一组预定义的捆绑包来利用多路复用增益来最大化实用性。我们制定了一个优化问题，并从理论上分析了约束的不规则性和问题的难度。我们开发了一种新颖的基于强化学习（RL）的切片分配解决方案。最后，我们进行了广泛的数据驱动的模拟实验。数值结果表明，所提出的解决方案可以有效地解决网络切片分配问题，并且比最佳基线算法具有更高的实用性。