This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LCOMM.2020.2977906, IEEE Communications Letters JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 1 Airtime-based Resource Allocation Modeling for Network Slicing in IEEE 802.11 RANs P. H. Isolani, N. Cardona, C. Donato, G. A. Perez,´ J. M. Marquez-Barja, L. Z. Granville, and S. Latre´ Abstract—In this letter, we propose an airtime-based Resource Nowadays, advanced 5G services and applications have Allocation (RA) model for network slicing in IEEE 802.11 stringent Quality of Service (QoS) requirements, which make Radio Access Networks (RANs). We formulate this problem as a the RA problem even more complex. To cope with dynamic Quadratically Constrained Quadratic Program (QCQP), where the overall queueing delay of the system is minimized while and unstable wireless environments and to meet QoS re- strict Ultra-Reliable Low Latency Communication (URLLC) quirements, network slices might be instantiated, modified, constraints are respected. We evaluated our model using three and terminated at runtime. Therefore, such slices have to different solvers where the optimal and feasible sets of airtime be properly and dynamically managed. To the best of our configurations were computed. We also validated our model with knowledge, this is the first work to exploit the flexibility of experimentation in real hardware. Our results show that the solution time for computing optimal and feasible configurations slice airtime allocation considering the IEEE 802.11 RAN re- vary according to the slice’s demand distribution and the number source availability and the stringent latency Key Performance of slices to be allocated. Our findings support the need for Indicator (KPI) related to URLLC. We use a scheduling policy precise RA over IEEE 802.11 RANs and present the limitations able to assign different airtime configurations to each slice, of performing such optimizations at runtime. named Airtime Deficit Weighted Round Robin (ADWRR) [5]. Index Terms—Resource Allocation, Network Slicing, Quality Thus, each slice can be configured according to their priority. of Service (QoS), URLLC, IEEE 802.11 RANs. In this letter, we focus on an RA model that optimizes the overall queueing delay of the system while QoS constraints I. INTRODUCTION are respected. Given that flows are distributed among slices HE fifth generation of mobile networks (5G) aims to that, in this case, represent abstractions of queues, we model T enable applications to run with lower latency, more reli- this problem using concepts of Queuing Theory. For that, ability, massive connectivity, and improved energy efficiency. we propose a Quadratically Constrained Quadratic Program The ITU Radiocommunication Sector (ITU-R) has defined (QCQP). We evaluated our model using Advanced Process three main uses for 5G: Enhanced Mobile Broadband (eMBB), OPTimizer (APOPT), Interior Point OPTimizer (IPOPT), and Ultra-Reliable Low Latency Communication (URLLC), and Z3 [6] solvers. We then validated our model with experi- Massive Machine Type Communications (mMTC); these are mentation in real hardware. Our results show that runtime envisioned to be deployed in the coming years. Among the optimization is limited by the demand distribution and the stringent requirements, low latency is seen as crucial and number of slices to be allocated. Our findings support the URLLC as the key enabler in this new age of connectivity. need for precise RA over IEEE 802.11 RANs and present Network slices, besides operating independently from one the limitations of performing such optimizations at runtime. another, provide networking resource and traffic isolation among users and services [1]. In the literature, many proposals II. IEEE RAN-RA PROBLEM FORMULATION focus on network slicing for IEEE 802.11 networks [2]. Network slicing is being used to address the Resource Allo- The IEEE 802.11 RAN consists of a set Access Points (APs) cation (RA) problem in IEEE 802.11 Radio Access Networks responsible to deliver data from different services to several (RANs), providing the required bandwidth and allocating users in the network, i.e., Stations (STAs). Each AP has re- resources accordingly [3]–[5]. However, in addition to the sources to be shared and therefore has to be properly managed. required bandwidth, End-to-End (E2E) latency and reliability Multiple tenants (i.e., virtual operators or service providers) should be considered as well. Moreover, deciding how to share from the same infrastructure and have their specific efficiently manage slice resources is still an open issue. Service Level Agreements (SLAs). These SLAs are translated into QoS requirements that the network has to support (e.g., P. H. Isolani, C. Donato, and S. Latre´ are with the IDLab, Department of minimum throughput, maximum allowed E2E latency, and Computer Science, University of Antwerp - imec, Antwerp, Belgium (e-mail: fpedro.isolani, carlos.donato, [email protected]). acceptable packet loss ratio). To meet such requirements, we N. Cardona and J. M. Marquez-Barja are with the IDLab, Department of propose the use of network slicing. According to Richart et Applied Engineering, University of Antwerp - imec, Antwerp, Belgium (e- al. [3], there are two variants of network slicing. The first mail: fnelson.cardona-cardenas, [email protected]) N. Cardona is also with Smart Networks and Services, Fondazione Bruno abstracts the different services and ensures QoS within them, Kessler Via Sommarive 18, 38123, Trento, Italy (e-mail: [email protected]). which is referred to as Quality of Service Slicing (QoSS). G. A. Perez´ is with the FOTS lab, Department of Computer Science, Uni- The second defines slices as the traditional idea of network versity of Antwerp, Belgium (e-mail: [email protected]) L. Z. Granville is with the Computer Networks Group, Federal University virtualization, where a precise subset of network resources is of Rio Grande do Sul, Brazil (e-mail: [email protected]). allocated to each tenant and full control is provided, which 1089-7798 (c) 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information. Authorized licensed use limited to: Universiteit Antwerpen. Downloaded on March 03,2020 at 07:39:09 UTC from IEEE Xplore. Restrictions apply. This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/LCOMM.2020.2977906, IEEE Communications Letters JOURNAL OF LATEX CLASS FILES, VOL. 14, NO. 8, AUGUST 2015 2 Network Services transmitting a frame and receiving the WiFi Acknowledgment Access Point 1 Traffic Rule Classifier (ACK) using the MCS RBEST. Note that Shortest Interframe Resource Block 1 Hypervisor Spacing (SIFS), DCF Interframe Spacing (DIFS), and the time RB2 Enqueuing Slice 1 Slice 1 Slice N for receiving the Acknowledgement TACK are considered. Given that network flows are distributed among slices and STA 1 STA 2 STA N STA 1 STA 2 STA N … … … …… slices represent abstractions of queues, we model this problem Buffering using Queuing Theory. Besides, since such flows’ arrivals are User Scheduler (Round-robin) User Scheduler (Round-robin) independent of one another and the traffic pattern is random, Slice Scheduler (ADWRR) we consider that flows’ arrivals follow a Poisson distribution. Scheduling To hardware queues Nonetheless, we acknowledge the inability to capture traffic burstiness in which characterizes some data traffic patterns. Fig. 1. Simplified slice queue structure along with the data traffic flow. However, our focus is to find the optimal assignment of Qs and validate it through real hardware experimentation. is called Infrastructure Sharing Slicing (ISS). As we focus on B. Mathematical Model QoS within a slice as being a service, we use the QoSS variant. The described network slicing RA problem can be ap- A. Problem Statement & Assumptions proached through optimization techniques. In this section, we present a QCQP approach for constraints. Since slices repre- To represent the minimum chunk of wireless resources sent queues and flow’s arrivals follow a Poisson distribution that can be assigned to a user, we use the Resource Block with exponential services, in Kendall’s notation, our model abstraction from Riggio et al. [7]. This abstraction defines a can be represented as sets of M/M/1. Given n services to WiFi interface at a given AP, identified by the network inter- be delivered by a resource block b, n slices are instantiated. face identifier (e.g., Medium Access Control (MAC) address), For each slice s in the set of slices Sb of a resource block operating channel (e.g., 1, 6, 11), and the type of channel (e.g., b, the maximum dequeuing rate µb , the dequeuing rate High Throughput (HT) 20MHz, Very High Throughput (VHT) MAX λs, the frame size FRMs , and the airtime As used for its 40MHz). Therefore, each resource block has its transmission size transmissions are provided. With such an input, the average capabilities, e.g., maximum dequeuing rate µb . MAX queueing delay W s of each slice can be calculated and the For each resource block, frames are first classified into the maximum average queueing delay W s can be ensured. different queues as slices, based on the definition of traffic QoS Besides respecting the service delivery rate and queueing delay rules (e.g., OpenFlow rules). Figure 1 depicts a simplified constraints, we aim to find the optimal assignment of Qs that queue structure along with the data traffic flow within a single minimizes the overall queueing delay. Therefore, AP. Thereafter, frames belonging to such slices/queues are X X dequeued following the ADWRR scheduling scheme [5]. With Minimize W s; (2) such a scheduling scheme, different portions of airtime, a.k.a., b2B s2Sb quantums, are allocated to each slice s in each transmission subject to the following set of constraints: round, denoted Qs.