Two-Level QoS-Oriented Downlink Packet Schedulers in LTE Networks: A Review Nasim Ferdosian and Mohamed Othman Department of Communication Technology and Network, Universiti Putra Malaysia, 43400 UPM, Serdang, Selangor D.E., Malaysia Email:
[email protected], mothman@ upm.edu.my Abstract. The Long Term Evolution (LTE) as a mobile broadband technology supports a wide domain of communication services with different requirements. Therefore scheduling of all flows from various applications by the same strategy of resource allocation would not be efficient. Accordingly it is necessary to design new scheduling algorithms by considering the Quality of Service (QoS) requirements defined for each application. In this regard, this paper will provide a brief overview of previously reported schedulers for QoS support in LTE cellular networks by taking into account the various important QoS parameters such as delay, packet loss ratio and data rate. The resource distribution problem in LTE networks and solutions from the numerous numbers of previous resource allocation approaches are outlined and compared through a comparative table. The future direction for solving these problems will be stated. Overall this study summarises the current state of knowledge on the QoS-oriented scheduling algorithms for LTE networks. Keywords: Long-term Evolution (LTE), scheduling, QoS, GBR and non-GBR.
1
Introduction
Mobile communication technologies are becoming constantly prevalent in our daily life with an expected increasing growth rate of mobile data higher than the fixed data. Nowadays mobile devices are able to support a wide range of different real-time and non-real time applications from call services to mobile Internet browsing, online TV, games, video-conference calls and so forth. According to the studies by Cisco [1] the monthly world’s mobile data traffic growth will reach 11.2 Exabyte by 2017 and video traffic will settle two-thirds of overall foreseen mobile data traffic. The Third Generation Partnership Project (3GPP) introduced the LTE [2] intending to design a system that can provide an efficient improvement in throughput over the older mobile standards and support multiple classes of QoS. In fact the most important novelty introduced by LTE technology is the enhanced support of QoS constraints among all its performance targets. With the aim of efficient support of current high variety of services, the efficient use of limited share bandwidth is essential. So the purpose of effective resource allocation strategies is crucial to meet the LTE targets (maximum spectral T. Herawan et al. (eds.), Proceedings of the First International Conference on Advanced Data and Information Engineering (DaEng-2013), Lecture Notes in Electrical Engineering 285, DOI: 10.1007/978-981-4585-18-7_67, Springer Science+Business Media Singapore 2014
597
598
Nasim Ferdosian and Mohamed Othman
efficiency, fairness, and QoS). However, it is typically impossible to accomplish all three intended goals at the same time [3] . Each factor can be supplied always at the cost of reducing another one. In this sense the main challenge is designing an allocation strategy to create a trade-off among the system performance and other desired targets of the network. Several studies have conceived the concept of QoS provisioning in LTE networks [4] . However, there have been no controlled studies to compare differences in QoS aware schedulers based on the QoS characteristics. In [5] the authors provided an inclusive overview of the common presented techniques in literature by representing a performance comparison of them with focus on all design aspects of the downlink packet scheduling in LTE. Another performance evaluation of representative scheduling strategies can be found in [6] limited to the uplink. The presented comparison of schedulers is in terms of their fairness, throughput and spectral efficiency. The authors in [7] considered interference mitigation and scheduling policy as two key design factors of LTE systems for achieving the performance goals and reviewed the scheduling methods proposed based on interference mitigation with the aim to increase the QoS of cell-edge nodes. In this paper we conducted a literature survey on the previously reported schedulers which have focused on support for QoS differentiation in LTE cellular networks, highlighting the important features of them, and classify them according to the QoS parameters considered as scheduling metrics for developing schedulers. In view of the fact that the most of the policy considerations are valid for both uplink and downlink resource allocation algorithms, therefore we focus on the downlink scheduling as the subject of our study. The rest of this paper is organized as follows. In section 2 we can overview the architecture of the LTE focusing on the issues of the resource sharing. Section 3 represents different characteristics of QoS in LTE networks pertain to design of resource sharing schedulers. In the fourth section we discuss approaches that are usually to use in this case and compare them in the fifth section. Finally in section 6, conclusions are drawn with particular attention to the new research directions.
2
Overview of Resource Allocation in LTE Downlink
Selecting an appropriate scheduling scheme is not standardized by the 3GPP specification for LTE [8] . Alternatively it is left to the vendors as an implementation decision, to adaptively configure and implement a well suitable algorithm according to the desired concerns of the system. It is worthy to mention that the responsibility of providing all these targets is up to the implementation of eNodeB residing in the MAC layer. The eNodeB assigns each active user a fraction of the total system bandwidth to share available resources among them by using a multiple access technique [9] . LTE downlink 3GPP adopted Orthogonal Frequency Division Multiple Access (OFDMA) as an access technique to accommodate a wide number of user equipment with different QoS application requirements and in different channel conditions. OFDMA allows multiple ac-
Two-Level QoS-Oriented Downlink Packet Schedulers
599
cesses by allocating disjoint selective collection of sub-carriers to each individual user to leverage multi user diversity and provide high scalability and robustness. The LTE radio resources are distributed in time and frequency domains. The LTE resource grid structure is displayed in 1. Each OFDMA frame is constructed of ten 1ms sub-frames in time domain and a sub-channel of 12 consecutive same size sub-carriers that cover 180 kHz of the frequency domain. The basic resource unit for mapping sub-carriers to active users is called Resource Block (RB). Each RB spans over a 0.5 ms time extent and one sub-channel.
Fig. 1. LTE radio resource grid structure
3
Quality of Service in LTE Networks
In the field of telecommunication network, the term Quality of Service indicates a measure of how efficient and reliable a network can fulfill a guaranteed level of satisfaction for its diverse services from real-time to non real-time services. Each application data flow is associated with a bearer. A bearer can be considered as a virtual connection established between Packet Data Network Gateway and User Equipment. Typically, the service bearers are categorized into two main classes, Guaranteed Bit Rate (GBR) and non-Guaranteed Bit Rate (non-GBR). On the other hand each bearer is characterized by a set of QoS attributes which is called QoS Class Identifier (QCI) [10] . QCI defines the QOS class to which the bearer belongs with the parameters associated with each data flow such as: Priority. The number given to each class of QoS which defines how emergency the bearer need to be resource allocated. Bearer Type. This parameter specifies the kind of the connected bearer: GBR bearer or non-GBR one. Packet Delay Budget (PDB). The maximum delay allowed for the transmission of queued packets. Packet Loss Rate. The maximum tolerable number of erroneous bits associated to a traffic flow of a given QCI.
600
Nasim Ferdosian and Mohamed Othman
The dedicated QCI label of bearer is one of the main factors that determine the behavior adopted by eNodeB for scheduling that particular type of bearer. The scheduler algorithms drive the specific allocation decision for each user based on a comparative metric per a specific RB. This metric is a scalar value which indicates a transmission priority for each user equipment founded on the information related to each stream of users. The related information is interpreted as the key parameters that construct the evaluating metric. Particularly in every kind of broadband wireless network there are a number of parameters that can be used to quantify the QoS application requirements and consequently enabling differentiation among flows carrying application data. These parameters consist of throughput, queue size, packet delay, packet loss rate and so on. The choice of these parameters for making priority metric is rather controversial, and there is no agreement about the number and kind of critical parameters which should be taken into account, however there are some issues we have to consider in choosing them: – Depends on the kind of traffic (real or non-real time) – The level of the complexity resulted from the created priority function
4
Downlink QoS-Oriented Scheduling Algorithms under Study
To date various scheduling techniques have been developed and introduced to serve downlink packet scheduling in LTE networks. In the following we go over the main points of the studied algorithms. The purpose of these scheduling algorithms is to distribute system resources among service flows by focusing on influential QoS factors in two main stages of time and frequency. We selected these specific methods due to their successful supporting of divers QoS requirements and their presented results. 4.1
Target Bit Rate
Authors in [11] introduced a QoS-oriented scheduler for Best Effort and Constant Bit Rate traffic focusing on the guaranteed bit rate requirements. To the end of controlling the signaling overhead and consequently decreasing the complexity, this solution is decoupled between time and frequency schedulers in the way that through each domain just a limited number of users are passed. The total users are grouped into two sets. First set contains users below their target bit rate and second set compromises remaining users. Then the prioritization of users within set 1 and 2 is done using common well-known (BET) Blind Equal Throughput and Proportional Fair (PF) approaches in time domain respectively. After selecting a number of candidate users by the Time Domain (TD) scheduler, they will be allocated resources in the frequency domain (FD) by using the PF scheme. The BET and PF metrics are expressed as: MBET = 1/(R[n])
(1)
Two-Level QoS-Oriented Downlink Packet Schedulers
¯ MP F = (D[n])/(R[n])
601
(2)
¯ where n is the user index, D[n] is the wideband throughput expected for the user n over all the bandwidth and R[n] is the past average throughput of user n which is updated every Transmission Time Interval (TTI) [12] . The main focus of this method is on improving total throughput along with considering guaranteed bit rate measurements as the only one QoS parameter. 4.2
Delay-throughput
A flexible QoS oriented scheduler based on two stages (time and frequency domains) proportional fair scheduling principle described in [13] for real time video traffic. The proposed algorithm considers arrival rate and head of line packet delay as QoS constraints. The proposed algorithm uses a metric which is a combined function of delay, throughput and Channel Quality Identifier (CQI) factors as follows: M = FD × FCQI × FT (3) where FD is a function QoS delay factors, FCQI is a function of CQI indicating channel state information of each user, and FT is factor of corresponding throughput calculation. In the time domain step which is the first phase of scheduling, all user equipments are sorted according to a new proposed ranking metric and in the next step, frequency domain, the actual RBs are allocated to the users with pending retransmissions, and then to the delay sensitive users. Finally the remaining RBs are given to the rest of priority users. This algorithm also cannot be considered as a strong QoS provisioning scheduler because of the ignoring other QoS factors such as the minimum data-rate requirements. 4.3
QCI-throughput
The authors in [14] applied a self-optimization method to the LTE network scheduler in response to the active changes of network conditions and traffic over time and proposed an Optimized-Service Aware (OSA) scheduler. To simplify the complexity of the resource allocation procedure, it has been partitioned into three separate stages, QoS classes identified classification, time domain and frequency domain scheduling. At the first step each bearer is classified into different QoS classes based on its CQI factor. Then the TD scheduler prioritizes the classified bearers according to their QoS requirements and categorizes them into separate prioritized candidate bearers: GBR and non-GBR. GBR bearers typically carry real time applications sensitive to delay and need to be served with a guaranteed bit rate. OSA algorithm sorts each GBR bearer according to the Head of Line (HOL) packet delay in the buffer of related bearer. On the other hand the nonGBR bearer list is ordered according to the following priority metric: ¯ MOSA = (D[n])/(θ[n])W QoS
(4)
602
Nasim Ferdosian and Mohamed Othman
where θ[n] is the normalized average channel condition estimate of bearer n and WQoS is the QoS weight. Two created sorted candidate groups are passed through the FD scheduler to be assigned optimal spectrum. The FD scheduler allocates the best RB to the highest GBR priority bearer. After giving enough resources to all GBR bearers, if any RB is still remained, FD scheduler assigns them to non-GBR bearers. The OSA algorithm can be demonstrated to be unsuitable for dealing with bounded losses as another factor of QoS support. 4.4
Combination of Multiple QoS Parameters
The ranking function of traditional scheduling algorithms which are only based on the queue’s priority, ignoring other metrics, would impose a lack of sufficient intellect over the resource allocation process. In response to this challenging problem, the authors in [15] introduced a new TD scheduling algorithm with emphasis in overload states. This overload state allocation algorithm supports QoS constraints by ordering the bearers using a ranking function of multiple metrics including priority, loss, delay, and queue depth. A priority value is considered for each metric, in order to make emphasis over any critical measurement. This ranking function is a combination of normalized metrics, individually multiplied in their corresponding priority as follows: X MKnapsack = Ri (5) i
where Ri indicates normalized prioritized metric i and i can be any subset of afore mentioned QoS measurement parameters such as packet loss and delay. Eventually an FD scheduler as an open complementary option can be selected among the existed schedulers to assign appropriate radio RBs to the optimal set of data packets for the sake of throughput.
5
Discussion
Providing the required QoS is vital to deliver a good user experience over the mobile Internet. The notion of QoS is becoming even more important as device capabilities have revealed the desire for consumers to use more rich media content such as video. The new scheduling services enhance the capabilities in providing the required QoS for next-generation mobile Internet applications. Through this study, we concisely explained a selection of downlink resource allocation approaches and explored their ability to efficiently support diverse QoS requirements. Different QoS mechanisms studied in this paper followed a network initiated QoS control based on the GBR and non-GBR bearers, which is a class-based packet forwarding treatment for delivering real-time and non-realtime traffics. Likewise they followed a two level framework where the resource allocation procedure is divided between TD and FD schedulers. Accordingly this separation can make it easier to optimize each step of scheduling independently and result in significant reduction of computational complexity.
Two-Level QoS-Oriented Downlink Packet Schedulers
603
In order to comply with the QoS requirements, the aforementioned algorithms present a metric based on a single or multiple QoS factors. The knapsack algorithm introduced an adaptive ranking function that can be generalized and make possible adding any desired traffic metric. It also allows the operator to define the critical level of each metric by assigning a specific priority level. Table 1. Comparison of studied QOS-oriented scheduling algorithms Scheduler Name
QoS Parameters
[11]
Target Bit Rate
[13] OSA [14] Knapsack [15]
Type of Traffic
Best Effort and Constant Bit Rate traffic Delay Video traffic QCI Real time and non- real time Delay, Loss, Queue Depth, Pri- VoIP and data connections ority, MBR/AMBR
On the other hand, from the viewpoint of the performance enhancement target of LTE systems, the studied algorithms take into account either the expected or achievable throughput of each user, except the last algorithm which responses to the need of high performance by mapping the LTE resource allocation scenario to the fractional knapsack problem and solving it by utilizing a greedy algorithm. Intuitively there is no need to use channel feedback for calculating user throughput. Therefore as a result, it would cause a reduced signaling overhead. Table 1 demonstrates the comparison of these different downlink scheduling.
6
Conclusion
Despite the huge amount of resource allocation approaches proposed so far, it is difficult to have a simple and fair weighting policy that can meet all QoS requirements of all connections in a time-varying wireless network. Therefore it is needed to introduce a principle of metric decoupling between time domain and frequency domain schedulers that is fundamental for maximizing throughput control. In this regards, in order to obtain a deep understanding and conceptual view of the QoS based scheduling concept, we discussed and classified the most interested and relevant techniques, from our point of view, presented in the literature according to their leveraging parameters of QoS. By making the allocation decision depending on the QoS attributes of each flow, the service provider significantly can increase the system revenue by providing their customers with an integrated variety of service plans. Consequently, this comparative study would reveal the need for resource allocation schemes in the way that obtain all LTE concerned targets simultaneously. Therefore channel aware approaches must be used with QoS aware strategies to provide a good balance between multi-QoS provisioning to support mixes of real-time/non-real-time traffic and system performance maximization.
604
Nasim Ferdosian and Mohamed Othman
As another future direction of research we can investigate the reasons of the issue that in practice the proposed schedulers so far are almost never implemented by switching and router manufacturers and QoS services are virtually never implemented and billed by Internet Service Providers. Acknowledgments. This work has been supported by the Malaysian Ministry of High Education under the Fundamental Research Grant Scheme FRGS/1/11/ SG/UPM/01/1.
References 1. Cisco. 2012. Cisco Visual Networking Index: Global Mobile Data Traf?cForecast Update,2011-2016, http://www.cisco.com/en/US/solutions/collateral/ns341/ ns525/ns537/ns705/ns827/white-paper-c11-520862.pdf. 2. 3GPP, Tech. Specif. Group Radio Access Network-Requirements for Evolved UTRA (E-UTRA) and Evolved UTRAN (E-UTRAN), 3GPP TS 25.913 V9.0, (2009) 3. Gui X., Ng T. S.: Performance of Asynchronous Orthogonal Multicarrier CDMA System in a Frequency Selective Fading Channel, IEEE Transactions on Communications, 47(7), 1084-1091, (1999) 4. Ekstrom, H.: QoS control in the 3GPP evolved packet system. IEEE Commun. 47, 76-83, (2009) 5. Capozzi F., Piro G., Grieco L. A., Boggia G., Camarda P.: Downlink Packet Scheduling in LTE Cellular Networks: Key Design Issues and a Survey. IEEE Commun. Surveys and Tutorials. 15(2), 678 - 700, (2012) 6. Elgazzar K., Salah M., Taha A. M., Hassanein H.: Comparing Uplink Schedulers for LTE. In: 6th International Wireless Communications and Mobile Computing Conference. pp. 189-193, ACM, (2010) 7. Kwan, R., Leung, C. : A survey of scheduling and interference mitigation in LTE. J. Elect. Comput. Eng. 2010, 1-10, (2010). 8. 3GPP. Tech. Specif. General Packet Radio Service (GPRS) enhancements for Evolved Universal Terrestrial Radio Access Network (E-UTRAN) access, TS 23.401 v9.4.0, (2010) 9. Ghosh A., Zhang J., Andrews J., Muhamed R.: Fundamentals of LTE. Prentice Hall, (2010) 10. 3GPP, Tech. Specif. Group Services and System Aspects - Policy and charging control architecture (Release 9). 3GPP TS 23.203, V9.4.0, (2010) 11. Monghal G., Pedersen K. I., Kovacs I. Z., Mogensen P. E.: QoS oriented time and frequency domain packet schedulers for the UTRAN long term evolution. In: Proc. of IEEE Veh. Tech. Conf., VTC-Spring, Marina Bay, Singapore, (2008) 12. Xalali R., Padovani R., Pankaj R.: Data throughput of CDMA-HDR a high efficiency-high data rate personal communication wireless system.In: Vehicular Technology Conference Proceedings,(VTC Tokyo). Vol.3,pp.1854-1858,IEEE,(2000) 13. Nonchev S., Valkama M.: QoS-oriented packet scheduling for efficient video support in OFDMA-based packet radio systems. In: Multiple Access Communications. pp. 168-180. Springer Berlin Heidelberg, (2011) 14. Zaki Y., Weerawardane T., Gorg C., Timm-Giel A.: Multi-qos-aware fair scheduling for lte. In: Vehicular Technology Conference (VTC Spring). pp.1-5. IEEE,(2011) 15. Brehm M., Prakash R.: Overload-state downlink resource allocation in LTE MAC layer. Wireless Networks Springer. 1-19, (2012)
