Elsevier

Computer Networks

Volume 94, 15 January 2016, Pages 344-359
Computer Networks

Service-differentiated downlink flow scheduling to support QoS in long term evolution

https://doi.org/10.1016/j.comnet.2015.11.002Get rights and content

Abstract

The growing demand of network services, especially for broadband downlink communication, has triggered the evolution of cellular systems. Recently, 3GPP continuously works out the standard of long term evolution (LTE) in response to the oncoming 4G cellular system. LTE adopts the OFDMA technology for downlink communication, and divides the spectral resource into physical resource blocks (PRBs). However, the LTE flow scheduling problem, which asks how to allocate PRBs to downlink flows for communication, is not well addressed in the standard but has great impact on transmission efficiency. On the other hand, LTE categorizes flows into guaranteed bit rate (GBR) and non-GBR classes, where GBR flows usually have higher priorities and shorter delay constraint than non-GBR flows. Although various solutions have been proposed to the LTE flow scheduling problem, many of them generally aim at maximizing transmission efficiency or keeping fair transmission, which may starve non-GBR flows or cannot well support quality of service (QoS) for GBR flows. Therefore, the paper develops a service-differentiated downlink flow scheduling (S-DFS) algorithm by taking the aforementioned difference between flows into consideration. Except for increasing transmission efficiency, S-DFS has two major goals: (1) satisfying the QoS requirement of GBR flows, and (2) ensuring the data transmission of non-GBR flows. Specifically, S-DFS first deals out PRBs to flows according to their channel conditions and QoS class identifier (QCI) defined in LTE. Then, with the mechanism of resource reallocation, S-DFS can assign a dynamic amount of reallocatable PRBs to the flows whose packets are about to be dropped. Experimental results demonstrate that S-DFS can achieve higher LTE transmission efficiency. Furthermore, it not only reduces both dropping ratio and delay of GBR packets, but also improves data throughput of non-GBR flows.

Introduction

The telecommunications industry keeps growing and innovating over the past few decades [1]. Nowadays, people expect to freely access Internet anytime, anywhere through mobile devices, and they are consequently hungry for high-speed wireless communication. In response to the above need, Third Generation Partnership Project (3GPP) develops long term evolution (LTE) for the current (and possibly future) generation of cellular system, and today, LTE systems have been operated in many countries [2]. On the other hand, it becomes common to use network services with the demand of broadband downlink communication, for example, online gaming, multimedia streaming, and mobile TV [3]. According to the Cisco report in [4], Internet video streaming and downloads have taken a large share of global network bandwidth in 2014, and will grow to more than 80% of all consumer Internet traffics by 2019. These broadband downlink services obviously pose challenges in the design of LTE systems.

LTE employs the technology of orthogonal frequency division multiple access (OFDMA) for its downlink communication, and divides the spectral resource into two-dimensional array of physical resource blocks (PRBs). A PRB has the duration of 0.5 ms in the time domain and the length of 180 kHz in the frequency domain. PRBs are ‘non-sharable’ resources, so each PRB can be given to at most one user. The number of available PRBs depends on the downlink transmission bandwidth. LTE allows the bandwidth to be 1.4, 3, 5, 10, 15, or 20 MHz, which supports 6, 15, 25, 50, 75, or 100 PRBs, respectively. How to allocate PRBs to downlink flows for transmission will significantly affect LTE transmission efficiency [5], and we call it the LTE flow scheduling problem. However, 3GPP leaves this problem to the research and industrial communities.

Furthermore, to provide differential QoS for various services, LTE classifies flows into two categories: guaranteed bit rate (GBR) and non-GBR. GBR flows can support real-time services with strict delay constraints, such as voice over IP (VoIP), live-streaming video, and online games. Non-GBR flows are often used to provide non-real-time services with loose deadlines, for example, TCP-based applications. However, some existing solutions to the LTE flow scheduling problem try to improve the overall transmission efficiency on the cost of flows with bad channel conditions or small priorities, which could thus starve non-GBR flows. On the other hand, other solutions attempt to achieve fair transmissions among flows. However, when the PRBs are not sufficient to support all flows, they may not guarantee QoS for GBR flows.

Therefore, the aforementioned observations motivate us to develop the service-differentiated downlink flow scheduling (S-DFS) algorithm to efficiently solve the LTE flow scheduling problem. In addition to improving LTE transmission efficiency, our S-DFS algorithm has two primary objectives: 1) satisfying the QoS requirement of GBR flows and 2) ensuring the data transmission of non-GBR flows. Here, the first objective means that we have to meet the constraints of packet dropping and delay of GBR flows, while the second objective indicates that we should prevent non-GBR flows from starvation. To do so, our S-DFS algorithm considers not only the channel quality, queue status, and head-of-line (HOL) packet delay of each flow, but also its QoS class identifier (QCI), which is a scalar identifier defined by LTE to describe the QoS characteristics of GBR and non-GBR flows. In particular, the S-DFS algorithm allows flows to bid for PRBs by using their channel quality and QCI priority, so it can increase transmission efficiency and favor GBR flows. Then, the S-DFS algorithm searches for ‘reallocatable’ PRBs from the above resource allocation by using two adjustable parameters α and β, which avoids some flows occupying too many PRBs. Finally, these reallocatable PRBs are assigned to the flows with impending packet discard. This resource reallocation mechanism not only reduces potential packet dropping but also allows non-GBR flows to obtain necessary PRBs for transmission.

The contributions of this paper are summarized as follows:

  • We develop an efficient S-DFS solution to the LTE flow scheduling problem by differentiating flows based on their QoS characteristics.

  • S-DFS adopts a novel mechanism of resource reallocation to ‘redistribute’ a subset of resource to the flows in urgent need of PRBs.

  • By adjusting parameters α and β, S-DFS can flexibly fine tune the amount of resource given to different types of services.

  • Experimental results show that S-DFS not only reduces GBR packet dropping and delay, but also improves non-GBR data throughput, as compared with a number of popular LTE flow scheduling methods.

The remainder of this paper is organized as follows: The next section discusses related work. Section 3 presents the network model for the LTE flow scheduling problem, and Section 4 proposes our S-DFS algorithm to solve the problem. We then compare the performance of different LTE flow scheduling schemes in Section 5. Finally, Section 6 concludes this paper.

Section snippets

Related work

Flow scheduling has received a lot of research attention, and various methods are proposed for wireless networks. Both maximum throughput (MT) [6] and proportional fair (PF) [7] are two typical scheduling methods. MT seeks to maximize transmission efficiency by selecting the UE ui with the best channel quality: ui=argmaxi(ri(t)),where ri(t) is ui’s channel rate at the current time t. PF aims at supporting fair transmission among UEs. Thus, PF compares the current channel rate of each UE with

Network model

Fig. 1 presents the LTE network architecture, which consists of two major parts: LTE core network and LTE cells. The LTE core network is responsible for various jobs such as back-end management, billing, and connecting to external networks. On the other hand, user equipments (UEs) are actually served in LTE cells. We thus focus on the flow scheduling problem in a single LTE cell, where the central base station (called E-UTRAN Node B, or eNB for short) takes responsibility of distributing

The proposed S-DFS algorithm

In every TTI, the eNB will collect the information of CQI and queue status from each UE, and then execute the S-DFS algorithm to deal out the downlink PRBs to UEs’ flows. Fig. 2 presents the flowchart and parameters of our S-DFS algorithm, which contains the following four stages:

  • (1)

    Preliminary PRB allocation: We adopt a two-phase strategy to calculate the preliminary assignment of downlink PRBs. In the first phase, the eNB gives PRBs to UEs based on their channel quality (i.e., CQI values).

Performance evaluation

In this section, we evaluate the system performance of our S-DFS algorithm by adopting LTE-Sim [27], which is an open-source simulator developed for modeling LTE networks. Our experiments aim at investigating the flow scheduling result in an LTE macro-cell, and Table 2 presents the simulation parameters. In every TTI, the eNB is responsible for dealing out 100 PRBs to a number of UEs roaming in the macro-cell. We vary the number of UEs to observe the effect of different amount of traffic loads.

Conclusion

Conventional LTE flow scheduling schemes target at either improving the overall transmission efficiency or guaranteeing the fair transmission among flows. However, they do not differentiate flows by their traffic features, which could starve non-GBR flows or may not satisfy the QoS requirement of GBR flows. To solve these problems, we develop the S-DFS algorithm that refers to the QCI characteristics and various parameters of flows to allocate PRBs. Furthermore, with the help of the resource

Acknowledgments

You-Chiun Wang’s research is co-sponsored by the Ministry of Science and Technology under Grant No. MOST 104-2221-E-110-036-MY2 and MOST 104-2628-E-110-001-MY2, Taiwan.

You-Chiun Wang received the Ph.D. degree in computer science from the National Chiao-Tung University, Taiwan, in 2006. He is an assistant professor in the Department of Computer Science and Engineering, National Sun Yat-sen University, Taiwan. Dr. Wang served as a TPC member of more than 90 conferences such as IEEE INFOCOM, ICDCS, and WCNC. His research interests include broadband wireless communications, wireless LANs and PANs, sensor and actuator networks, and mobile and pervasive computing.

References (28)

  • W.K. Lai et al.

    QoS-aware downlink packet scheduling for LTE networks

    Comput. Netw.

    (2013)
  • E. Bertin et al.

    Evolution of Telecommunication Services

    (2013)
  • J.G. Andrews et al.

    What will 5G be?

    IEEE J. Sel. Areas in Commun.

    (2014)
  • H. Luo et al.

    Quality-driven cross-layer optimized video delivery over LTE

    IEEE Commun. Mag.

    (2010)
  • Cisco, Cisco visual networking index: forecast and methodology, 2014–2019 white paper, 2015,...
  • F. Capozzi et al.

    Downlink packet scheduling in LTE cellular networks: key design issues and a survey

    IEEE Commun. Surv. Tutor.

    (2013)
  • P. Kela et al.

    Dynamic packet scheduling performance in UTRA long term evolution downlink

    IEEE International Symposium on Wireless Pervasive Computing

    (2008)
  • H.J. Kushner et al.

    Convergence of proportional-fair sharing algorithms under general conditions

    IEEE Trans. Wirel. Commun.

    (2004)
  • M. Andrews et al.

    Providing quality of service over a shared wireless link

    IEEE Commun. Mag.

    (2001)
  • J.H. Rhee et al.

    Scheduling of real/non-real time services: adaptive EXP/PF algorithm

    IEEE Vehicular Technology Conference

    (2003)
  • B. Sadiq et al.

    Downlink scheduling for multiclass traffic in LTE

    EURASIP J. Wirel. Commun. Netw.

    (2009)
  • H. Luo et al.

    Quality-driven cross-layer optimized video delivery over LTE

    IEEE Commun. Mag.

    (2010)
  • G. Piro et al.

    Two-level downlink scheduling for real-time multimedia services in LTE networks

    IEEE Trans. Multimed.

    (2011)
  • K.J. Astrom et al.

    Computer-Controlled Systems: Theory and Design

    (1996)
  • Cited by (0)

    You-Chiun Wang received the Ph.D. degree in computer science from the National Chiao-Tung University, Taiwan, in 2006. He is an assistant professor in the Department of Computer Science and Engineering, National Sun Yat-sen University, Taiwan. Dr. Wang served as a TPC member of more than 90 conferences such as IEEE INFOCOM, ICDCS, and WCNC. His research interests include broadband wireless communications, wireless LANs and PANs, sensor and actuator networks, and mobile and pervasive computing. He has published more than 50 research papers and book chapters in these fields. He is a senior member of the IEEE and a member of the ACM.

    Song-Yun Hsieh received the M.S. degree in computer science from the National Sun Yat-sen University, Taiwan in 2015. His research interest focuses on 4G LTE systems.

    View full text