A Two-Timescale Approach for Network Slicing in C-RAN

Abstract

Network slicing is a promising technique for cloud radio access networks (C-RANs). It enables multiple tenants (i.e., service providers) to reserve resources from an infrastructure provider. However, users' mobility and traffic variation result in resource demand uncertainty for resource reservation. Meanwhile, the inaccurate channel state information (CSI) estimation may lead to difficulties in guaranteeing the quality of service (QoS). To this end, we propose a two-timescale resource management scheme for network slicing in C-RAN, aiming at maximizing the profit of a tenant, which is the difference between the revenue from its subscribers and the resource reservation cost. The proposed scheme is under a hierarchical control architecture which includes long timescale resource reservation for a slice and short timescale intra-slice resource allocation. To handle traffic variation, we utilize the statistics of users' traffic. Moreover, to guarantee the QoS under CSI uncertainty, we apply the uncertainty set of CSI for resource allocation among users. We formulate the profit maximization as a twostage stochastic programming problem. In this problem, long timescale resource reservation for a slice is performed in the first stage with only the statistical knowledge of users' traffic. Given the decision in the first stage, short timescale intra-slice resource allocation is performed in the second stage, which is adaptive to real-time user arrival and departure. To solve the problem, Abstract iv we first transform the stochastic programming problem into a deterministic optimization problem. We then introduce a maximum interference constraint and transform the QoS constraint under CSI uncertainty into linear matrix inequalities. We further apply semidefinite relaxation to transform the problem into a mixed integer nonconvex optimization problem, which can be solved by combining branch-and-bound and primal-relaxed dual techniques. Simulation results show that our proposed scheme can well adapt to traffic variation and CSI uncertainty. It obtains a higher profit when compared with several baseline schemes. v Lay Summary Network slicing is a promising technique for the fifth generation (5G) wireless systems. It allows multiple service providers to run on top of a shared physical network infrastructure. Meanwhile, cloud radio access network (C-RAN) is a centralized, cloud computing-based architecture for radio access networks to support various types of traffic demand in 5G wireless systems. However, implementing network slicing in C-RAN is faced with critical challenges due to time-varying network conditions and inaccurate knowledge of the conditions. In this thesis, to tackle the aforementioned challenges, we propose a dynamic resource management scheme for network slicing in C-RAN. Simulation results show that our proposed scheme can achieve a better performance when compared with several baseline schemes.

Service Provisioning in Sliced Cloud Radio Access Networks

“…Slice admission and bandwidth allocation were performed on a long timescale, and the operator generated beamformers on each short timescale. To achieve efficient resource utilization, some investigations have been conducted to design the resource reservation and intraslice resource allocation from the perspective of the system [34,35]. In [34], a network slicing game was proposed to make a resource reservation with the aim of maximizing its user utility.…”

A Resoure Allocation Framework for Network Slicing with Multi-service Coexistence

“…In this work, we consider slicing requests from eMBB and URLLC, two major application scenarios of 5G and beyond wireless network systems, and the bandwidth and computing resources required by URLLC slices and eMBB slices are denoted by b u , s u , b e and s e , and the computing resources leased from remote cloud servers is denoted by s c t . To cater the dynamic of the slice requests in practice, the timeslotted model is considered here, where time is divided into long time slots (LTSs) and short time slots (STSs) [12]. We denote LTS as L, and each LTS contains n STSs, denoted as L = (t 1 , t 2 , ..., t n ).…”

Resource Reservation in Backhaul and Radio Access Network with Uncertain User Demands

“…It is observed that although the expected outage is separable in r p k , variables are coupled in the first term of the objective function. Therefore, we substitute the global quadratic upperbound given in (21)…”

“…Therefore, the Lipschitz constant is |P k |. We now place the Lipschitz constant in (21) and find the upper-bound which is separable in r p k as follows:…”