IEEE TRANSACTIONS ONVEHICULAR TECHNOLOGY, VOL. 69, NO. 2, FEBRUARY 2020 2079 End-to-End Slicing With Optimized Communication and Computing Resource Allocation in Multi-Tenant 5G Systems Hsu-Tung Chien , Ying-Dar Lin , Fellow, IEEE, Chia-Lin Lai, and Chien-Ting Wang Abstract—Slicing is a key technology in 5G networks to provide scalability and flexibility in allocating computing and communi- cation resources among multiple tenants. Typically, 5G networks have a 2-tier architecture consisting of a central office and transport network in the upper tier and a multi-access edge and radio access network in the lower tier. The tenants which share the 2-tier archi- tecture typically have different service-dependent resource require- ments. This study proposes an algorithm, designated as Upper- tier First with Latency-bounded Over-provisioning Prevention (UFLOP), to adjust the capacity and traffic allocation in such a way as to minimize the “over-provisioning ratio” while still satisfying the latency constraints and Service Level Agreements (SLAs) of the tenants. The performance of UFLOP is evaluated experimentally with a real testbed on an end-to-end slicing framework using three typical 5G services, namely Enhanced Mobile Broadband (eMBB), Ultra-Reliable Low Latency (URLLC), and massive Machine Type Connection (mMTC). It is shown that UFLOP successfully deter- mines the critical traffic allocation ratio between the central office and the edge which achieves an over-provisioning ratio close to zero while still meeting the latency requirements. The results suggest optimal resource allocation ratios of 10:0, 1.5:8.5 and 7.8:2.2 for the eMBB, URLLC and mMTC applications, respectively. Fur- thermore, it is shown that the computing resource behaves as a bottleneck for the eMBB and mMTC services, while the communi- cation resource serves as a bottleneck for the URLLC service. Index Terms—Radio Access Network (RAN), Multi-access Edge Computing (MEC), slicing, computing resource, communication resource, virtualization, optimization. I. INTRODUCTION F OR infrastructure owners and service providers in multi- tenant 5G networks [1], slicing technology [2] provides an Manuscript received April 11, 2019; revised September 29, 2019; accepted December 8, 2019. Date of publication December 11, 2019; date of current version February 12, 2020. This work was supported in part by the H2020 Collaborative Europe/Taiwan Research Project 5G-CORAL under Grant 761586 and in part by the Duan Jin Research Project in the Institute of Information and Communications from the Industrial Technology Research Institute, Taiwan. The review of this article was coordinated by Dr. F. Tang. (Corresponding author: Hsu-Tung Chien.) H.-T. Chien and Y.-D. Lin are with the Department of Computer Science, National Chiao Tung University, Hsinchu 300, Taiwan (e-mail: hsutung@cs.nctu.edu.tw; ydlin@cs.nctu.edu.tw). C.-L. Lai was with the Information and Communications Research Laborato- ries, Industrial Technology Research Institute, Hsinchu 300, Taiwan. She is now with MediaTek, Inc., Hsinchu 30078, Taiwan (e-mail: chia-linlai@itri.org.tw). C.-T. Wang is with the Graduate Degree Program of Network and Information Systems, National Chiao Tung University, Hsinchu 300, Taiwan and Academia Sinica, Taipei 115, Taiwan (e-mail: ctwang@cs.nctu.edu.tw). Digital Object Identifier 10.1109/TVT.2019.2959193 essential tool for meeting the diverse deployment requirements of the different tenants. Among the various service types de- ployed by 5G tenants, Enhanced Mobile Broadband (eMBB) [3], Ultra-Reliable Low Latency (URLLC) [4], and massive Machine Type Connection (mMTC) [5] are among the most common. eMBB services usually require high bandwidth and a large computational power, and may have either a tight or loose End-to-End (E2E) latency constraint, depending on the particular service involved, e.g., video streaming, Augmented Reality (AR), or Virtual Reality (VR). By contrast, URLLC services, such as Vehicle-to-everything (V2X), demand high reliability connections with E2E latency constraints as tight as 1 ms [4], [6]. Finally, mMTC services (e.g., Internet of Things (IoT) gateways) involve a sudden and massive burst of connections, and typically have a loose latency constraint (e.g., larger than 50 ms to a second). In multi-tenant 5G networks, it is frequently necessary to support these different service types at the same time. As a result, some form of resource isolation mechanism is required to ensure that a sufficient amount of resource is retained for each service. Furthermore, an efficient means of allocating the isolated resources in the 5G network in such a way as to guarantee the Quality of Service (QoS) requirements of the tenants is also required. The resource isolation problem is generally solved by slicing; a technology which virtualizes the physical resources of the net- work and then partitions these resources into isolated instances, such as virtual machines and containers, for service deployment. In a previous study [7], the present group proposed a Joint Edge and Central Resource Slicer (JECRS) mechanism for slicing the network and computing resources in 2-tier 5G architectures consisting of a central office and a transport network in the upper tier and a Radio Access Network (RAN) [8] and multi-access mobile edge (MEC) [9], [10] in the lower tier. The experimental results showed that the slicing isolation effect ratio was equal to 1; indicating that the slices were independent of one another. Since the resources at the edge are typically much rarer than those at the central office, the user traffic should ideally be routed to the central office for processing (providing that the latency constraint allows). In [7], the present group examined the latency requirement stuffocation ratio of three major 5G services types and their representative applications for various fixed configured ratios of the resource allocation distribution between the edge and central office. The results suggest a Central to Edge resource 0018-9545 © 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See https://www.ieee.org/publications/rights/index.html for more information. Authorized licensed use limited to: National Chiao Tung Univ.. Downloaded on April 03,2020 at 09:30:34 UTC from IEEE Xplore. Restrictions apply.