Performance Evaluation Of Various WiMAX Schedulers Under Triple Play Services

Transcription

1 Performance Evaluation Of Various WiMAX Schedulers Under Triple Play Services Anju Lata Yadav Dr P.D.Vyavahare Shri G.S.Institute of Technology and Science, Indore (MP) Dr. P.P.Bansod Abstract IEEE based Worldwide interoperability for Microwave Access (WiMAX) is a Broadband Wireless Access (BWA) technology that supports various real time multimedia applications (voice and video) as well as non real time data applications such as file transfer and . The Quality Of Service (QoS) requirements of these triple play services are satisfied by various MAC layer schedulers that differentiate real time and non real time applications by division of into five service classes, admission control and scheduling the order of packet transmission. In this paper major homogeneous scheduling algorithms are evaluated for QoS metrics (average throughput, average delay, and average jitter) and fairness to discover their suitability and their shortcomings for particular triple play service application. Simulation results reveal that no single scheduling algorithm satisfies QoS requirements of all applications. Therefore designing an optimal WiMAX scheduling algorithm is still an active area of research. Index Terms: WiMAX Scheduling algorithm, Triple play Services, QoS, Fairness in WiMAX. I INTRODUCTION Next Generation Network is required to support diversified applications such as Video on Demand (VoD), Voice over IP (VoIP) and data on the same network through integration of these triple play services on a single network [1]. The converged network must also offer flawless quality and strict support for Service Level Agreement (SLA). QoS is defined in standard X.903 by ITU-T as a set of quality requirements on the collective behavior of one or more objects. QoS provides preferential treatment to applications by ensuring sufficient bandwidth, controlling delay and minimizing jitter and packet loss. Providing QoS simultaneously to the heterogeneous having different characteristics is a major challenge in wireless access networks. These services have varying QoS requirements and therefore are normally provisioned individually. The standard IEEE has not defined the provisioning of these services. WiMAX has the potential to deliver these services satisfying variable QoS requirements. IEEE standard was designed with a variety of types in mind, to handle the requirements of high data rate applications, such as VoIP and video or audio streaming, as well as low data rate applications, such as web surfing and handle extremely busty over the internet simultaneously [1].QoS requirements of applications attached to WiMAX service classes is outlined in table I [2]. It is observed that an acceptable delay for VoIP is 150 ms and more than 200 ms delay results in noticeable impaired voice quality. WiMAX shares the wireless media by two types of operational modes. They are Point To Multipoint (PMP) and mesh topology. In PMP mode, Subscriber Stations (SSs) can communicate with each other through centralized Base Station (BS) and in mesh network, either SSs communicate directly or they help each other to relay packets between source and destination SS. IEEE standard has specified five scheduling class services to support diverse applications along with their bandwidth request/grant mechanisms. They are Unsolicited Grant Service (UGS), extended real time polling service (ertps), real time polling service (rtps) non real time polling service (nrtps) and Best Effort (BE) [3]. The allocation of wireless resources for QoS provisioning is not specified in the standard. The standard provisions for packet scheduler which differentiates between different types of packets. Traditional First Come First Served (FCFS) packet scheduler policy does not differentiate between packets belonging to different applications. Packet classification is, therefore, usually achieved by assignment of different packets to different queues. Scheduler then chooses the next packet to be dequeued from the available multi queues and forwards it to output link [4]. Various proposals of scheduling algorithms for WiMAX are based on legacy scheduling algorithms designed specifically for WiMAX and have been changed as per the specifications of standard. These proposals have been broadly classified as homogeneous, hybrid and opportunistic scheduling algorithms [5]. Homogenous scheduling algorithms such as Weighted Fair Queuing (WFQ), Earliest Deadline First (EDF), Weighted Round Robin (WRR), Round Robin (RR), and Strict Priority (SP) are the legacy scheduling algorithms aimed to satisfy the QoS requirements of various applications. Hybrid scheduling algorithms combine legacy scheduling algorithms to make an assorted scheduling algorithm. Opportunistic scheduling algorithms consider the variability in channel conditions. Homogeneous scheduling algorithms were originally designed for wired network. However, they are also being used for unpredictable wireless network. The suitability of such scheduling algorithms in satisfying the QoS requirements of triple play services voice, video and data in a fixed and mobile WiMAX network has been a topic of research investigations for the last few years [6]. The idea behind this research paper is providing QoS to applications of triple play services in a fixed WiMAX network through /12/$ IEEE

2 uplink packet scheduling. The analysis and simulation study of these uplink homogeneous scheduling algorithms can be used to understand their usefulness for various applications. Further such a comparison of various scheduling algorithms can be used for modifications in scheduling algorithms to design new algorithms which are more suitable in WiMAX environment. The rest of the paper is organized as follows. In section II scheduling architecture and homogeneous scheduling algorithms in WiMAX Network are discussed. Section III describes the simulation framework, results and their analysis. Section IV summarizes the results and the future research directions. Table I: QoS Requirements of Applications in WiMAX [2] Service Class Applications Bandwidt h (kbps) Latency (msec) Jitter (msec) UGS, rtps Multiplayer Interactive game (50-85) (<150) (<100) UGS, Ertps VoIP and Video (4-384) (<150) (<50) Conference Rtps Streaming Media (5-2000) N/A (<100) Nrtps, Web Browsing Moderate N/A N/A BE Nrtps, BE Media content downloads ( ) >2000 N/A N/A II SCHEDULING ALGORITHMS Packet scheduler decides about contention for bandwidth in a shared wireless media. There are three types of scheduling in IEEE , downlink and uplink scheduling performed at BS [7] and uplink scheduling performed at SS. The DownLink Base Station (DL-BS) scheduler decides queue number to be served and the number of data units to be transmited to SS. Their number depends on QoS parameters and channel state condition. The UpLink Base Station (UL- BS) scheduler decides the allocation of channel among users along with transmission order. The bandwidth allocation depends on bandwidth requests from SSs and the required QoS parameters. Designing an uplink scheduler is more challenging than downlink scheduler as complete information about queue size is not available with it. Bandwidth provided by UL-BS scheduler can be allocated per connection, known as Grant Per Connection (GPC) or per station known as Grant Per Subscriber Station (GPSS). GPSS is recommended by latest version of standard to transfer the burden of bandwidth allocation for each connection to SS scheduler, thereby relieving BS. The allocated resource is in the units of number of slots. These slots are then mapped into a number of sub channels which are a group of multiple physical subcarriers [8]. The important task of scheduling algorithm is to satisfy the QoS requirements of each user while efficiently utilizing the available bandwidth for the uplink. The uplink scheduling algorithm works along with Call Admission Control (CAC) and decides the next packet that is to be transmitted based on its policy to satisfy QoS requirements of existing and new users (shown in fig.1). CAC algorithm ensures that a connection is accepted into the network only if its QoS requirements can be satisfied without deteriorating the performance of existing connections in the network [7]. Figure: 1 IEEE QoS Architecture There are certain desirable characteristics to be possessed by the scheduler. They are, to meet QoS requirements of five service classes, maximize system goodput, maintain fairness, link utilization, minimize power consumption, conserve power in mobile devices, simplicity with reduced complexity, ease in system scalability, flexibility of deployment in different networks or technology upgradation with minimum changes, protection of users who abide by Service Level Agreement (SLA) from those producing unpredictable fluctuations in the network by not following SLA [ 5, 7]. Homogeneous scheduling algorithms or channel unaware scheduling algorithms, do not consider the variations in channel condition. Such algorithms, being simple, can satisfy QoS requirements of various applications in WiMAX. These schedulers provide QoS requirements among five service classes satisfying delay and throughput constraints. Jitter is also one of the QoS parameter. However, none of these algorithms can guarantee upper limit on jitter [7]. Homogeneous scheduling algorithms considered in this paper are RR, WRR, SP and WFQ. These algorithms are discussed next. (a)round Robin: It is a very basic algorithm with a complexity of O (1) which serves each queue one after the other and dequeues one packet from each queue. RR provides allocations to even idle connections. RR is suitable for fixed sized packets. However, the packets of variable size such as MPEG video suffer from fairness point of view. (b)weighted Round Robin: It is similar to RR. WRR visits the queue in a round robin fashion. To support different data rates and service classes, weights are assigned to each queue

3 as per the throughput and maximum delay restrictions of each service class and packets are dequeued in proportion to their weights. The complexity of WRR is same as that of RR. The algorithm is not suitable for variable sized packets as the assignment of bandwidth is achieved according to weights. Unlike RR bandwidth allocations are not made to idle connections. Weights can be assigned to service classes in proportion to their priority such as higher weight to rtps as compared to nrtps and BE. Number of parameters can be utilized to assign weights to each queue. They are the amount of delay, jitter or packet loss [9]. The weights can be assigned with the help of QoS parameter, Minimum Reserved Traffic Rate (MRTR), as given by equation 1, specified by SS upon admission into the network. [5] Where W i is the weight of i th SS, n= number of SS. (c)strict Priority: It places the packets in different priority queues depending on QoS classes. priority packets are scheduled first till the queue is empty. If higher priority queues have more than lower priority, lower priority suffers from bandwidth starvation. SP scheduling algorithm is utilized by Internet Service Providers (ISP) to carry VoIP to guarantee the requisite QoS guarantee by giving them highest priority. In a WiMAX network priority given to each service is in the following order UGS>ertps>rtps>nrtps>BE. SP is suitable for serial lines with low bandwidth having static configuration [10]. (d)weighted Fair Queuing: In WFQ, flows with different bandwidth requirements are supported by assigning different weights to queue which is proportional to bandwidth. As appropriate weight is assigned to packets, variable length packets are supported so that flows of larger packet size are not allocated more bandwidth. WFQ is an approximation to theoretical scheduler Generalized Processor Sharing (GPS) in which a packet divided into bits are transmitted in weighted round robin fashion at the start of each queue [11]. Finish time of a packet is the time packet would have been reassembled at the assembler, if transmitted by GPS algorithm. The finish time depends on the weight assigned to each packet, the size of the packet and the capacity of uplink channel. In WFQ finish time is assigned to packets which are selected in the ascending order of finish time. The finish time does not represent the actual transmission time, rather it is the number assigned to each packet that signify the transmission order of a packet. The complexity of WFQ is O(N) and its implementation is complex as compared to other scheduling algorithms,as scheduler has to maintain states of each service class and scan the states of each packet on their arrival and departure. The scalability of WFQ is affected by the complexity when it is required to support large number of service classes. III SIMULATION ENVIRONMENT AND RESULTS WiMAX homogeneous scheduling algorithms, SP, RR,WRR,WFQ, Self clocked Fair(SCF) Queuing, Diff- Serv(DS) are evaluated in [12] to reflect the effect of queue size and number of queues on their performance. Homogeneous, hybrid and opportunistic scheduling algorithms are evaluated for various applications VoIP, video streaming FTP and HTTP to study the effect of multiplexing various types on intraclass and interclass fairness, average delay, packet loss and the average throughput. WRR as uplink scheduler, Deficit Round Robin (DRR) as downlink scheduler at BS and DRR can be applied as SS scheduler [13]. The scheduling is controlled by BS and so scheduling algorithms employed for wired network can provide QoS guarantee. Mobile WiMAX s performance is affected by different factors such as load, mobility and coverage with different types. The performance metrics are affected more by the load on the BS as compared to mobility [14]. The assessment of major uplink homogeneous scheduling algorithms have been done using Qulanet 4.5 simulator to investigate the best and most efficient uplink scheduling algorithm in WiMAX network, suitable for a particular application. The WiMAX network selected for investigation is a single cell PMP consisting of one BS and a number of SS varying from 10 to 25. TABLE II SIMULATION PARAMETERS Parameter Value Physical Layer Wireless MAN OFDM Simulation Grid size 1000m x 1000m Frame structure TDD Channel Bandwidth 20 MHz BS transmit Power 30 dbm Simulation time 50 sec DL:UL frame ratio 1:1 Node Placement Random Ertps:rtps:nrtps:BE 2:2:1:1 Statistics are collected for performance metrics throughput, average delay, average jitter and fairness index. Intraclass fairness and interclass fairness are two types of fairness. Intra class fairness is measured between users of same class and is calculated using Min-Max index as defined in equation 2. Interclass fairness is measured between all users and is calculated using Jain s Fairness Index (JFI) as given in equation 3[15]. Where x min and x max denote the minimum and maximum average throughput in a class, and x i denotes the average throughput for i th SS. Implemention of scheduling of triple play services for different sources is achieved by attaching them to each service class. VoIP can be best served with either UGS or ertps. Although UGS provides lower packet loss but it does

4 not utilize the network resources effectively with variable bit rate. Rtps is designed for audio streaming and streaming video. Rtps can request the bandwidth periodically as the fixed bandwidth may not be utilized [16]. VoIP is modeled for SSs of ertps class, streaming video for SSs of rtps class, FTP for SSs of nrtps and BE service class. VoIP uses H.323 call signalling with average talking time of 20 seconds. The video model consists of two data of bit rate Mbps and Mbps. FTP model consists of constant packet size of 1000 bytes. Figures 2 and 3 show the variation of average delay and average jitter respectively with the increase in number of SS for VoIP served by ertps for the four homogeneous scheduling algorithms WRR, WFQ, RR and SP. Average delay is highest for SP and delay response of WRR, WFQ and RR is almost same. When the number of SS increases, average jitter increases sharply for WRR and RR, while it is constant for WFQ. As WFQ assigns weight according to bandwidth requirements which is the highest for ertps, it results into lower average delay and average jitter for VoIP. Ertps is designed to support real-time service flow that generates variable size data packets on a periodic basis, such as VoIP services with silence suppression. WFQ is superior as compared to WRR in the presence of variable sized packets and therefore when VoIP is scheduled by WFQ, least average delay and jitter is obtained. For the application VoIP, though SP also provides low jitter but it is unfair to lower priority data which suffers from starvation, if higher priority real time is in access. Throughput is not a requisite QoS parameter for real time applications, while delay and jitter play a vital role. Figures 4 and 5 show the variation of average delay and average jitter with the increase in number of SS, respectively, for streaming video served by rtps for the homogeneous scheduling algorithms WRR, WFQ, RR and SP. The average delay and average jitter increases sharply for all scheduling algorithms with the increase in the number of SS as there is an increase in bandwidth wasted by uplink burst preambles and increase in packet drop due to increase in load. Least average delay and least average jitter is obtained with scheduler WRR. As rtps always use a bandwidth request process for suitable size grants, it transports data more efficiently than UGS algorithm. However, this bandwidth request process always causes MAC overhead and additional access delay. Hence rtps algorithm has larger MAC overhead and access delay than UGS and ertps. WRR is suitable for fixed size packets and it provides higher weight to rtps as compared to non real time. Ertps is designed for variable size packets and so WRR provides low performance for VoIP served by ertps flow. Figure 6 shows the variation of average throughput with the increase in number of SS for data served by nrtps and BE for the homogeneous scheduling algorithms WRR, WFQ, RR and SP. Throughput decreases with the increase in the number of SS due to increase in network load and increase in wastage of bandwidth due to uplink burst preamble. Though RR provides lower throughput as compared to rest of the scheduling algorithms, it is suitable for data as it is most simple with complexity O (1). Figure 7 shows the variation of intraclass fairness, indicated by Min-Max index, with the increase in number of SS for data, for the homogeneous scheduling algorithms WRR, WFQ, RR and SP. er difference in maximum and minimum average throughput indicates higher intraclass fairness while higher difference indicates lower intraclass fairness. The intra class fairness for the data fluctuates as large number of SSs competes for the same amount of residual bandwidth after allocation to real time video and voice. Some SSs will receive large amount of bandwidth while others will have meager bandwidth. Since same amount of bandwidth is being shared by increased number of SS, there is a decrease of fairness among SSs. Figure 8 shows interclass fairness calculated by JFI. It has highest value for WRR and lowest for SP. WRR and WFQ algorithms distribute the bandwidth according to the weights, thus, the higher priority has a higher throughput than the lower priority. WRR overcomes the limitation of SP by allowing an access of bandwidth to lower priority by removal of at least one packet from each queue during each service round. Fig 2: Average Delay of VoIP served by ertps vs no.of SS Fig 3: Average jitter of VoIP served by ertps vs no. of SS Fig 4: Average Delay of streaming video served by rtps vs Number of SS

5 Fig 5: Average jitter of streaming video served by rtps vs Number of SS Fig 6: Average throughput of data served by Nrtp and BE vs number of SS Fig 7: Intra class fairness for data Vs Number of SS Fig 8: interclass fairness for scheduling algorithms IV CONCLUSION Increase of bandwidth intensive real time services in wireless network requires schedulers that can make provisions for the requisite amount of bandwidth for every application. WiMAX supports various real time and non real time applications through their classification in five service classes, their scheduling and call admission control. WiMAX MAC scheduling algorithms such as WRR,WFQ and SP provide better utilization of network resources. The investigation on behavior of uplink homogeneous scheduling algorithms namely WRR, WFQ, RR and SP have shown their suitability for different applications. Their comparison, as shown in table III, shows that the scheduling algorithm WRR is suitable for scheduling real time streaming video ; WFQ shows its suitability for VoIP and for data RR can be used. Results show that not a single scheduling algorithm is suitable for scheduling of all configurations of triple play services. The QoS requirements of all the applications cannot be satisfied by a single scheduling algorithm. An attempt to design a versatile scheduler which amalgamates the desirable features of major schedulers suitable for all applications is still an active field of investigation. TABLE III COMPARISON OF SCHEDULERS Algor ithm RR WRR SP WFQ Complexity (v.simple (simple) (simple) O(N) (complex Inter class fairness (Unfair) high (Unfair to variable pkt low (Unfair) Fair (Medium) REFERENCES Intraclass fairness (real/non real time) /medium /high /low / medium Traffic suitability Data Real time video Interactive voice VoIP [1] Jeffrey G. Andrews, Arunabha Ghosh, Rias Muhamed, Fundamentals of WiMAX: Understanding Broadband Wireless Networking, Prentice Hall, Edittion I, [2] WiMAX Forum, WiMAX System Evaluation Methodology, Version 2.1, July [3] IEEE : IEEE Standard for Local and Metropolitan area networks Part 16: Air interface for Fixed and Mobile broadband Wireless Access Systems, [4] Loutfi Nuaymi, Wi-MAX Technology for Broadband Wireless Access, John Wiley and Sons Ltd, Edition I, [5]Pratik Dhrona, NajahAbu Ali, Hossam Hassanein, A Performance Study of Scheduling Algorithms in Point to Multipoint WiMAX Networks, IEEE conference on Local Computer Networks(LCN), Montreal,October 2008, pp [6] Narbutaite, B. Dekeris, Triple Play services scheduling Performance Evaluation, Journal of Electronics and Electrical Engineering, 2008, No 6(86), pp [7] Chakchai So-In, Raj Jain, Abdel-Karim Tamimi, Scheduling in IEEE e Mobile WiMAX Networks: Key Issues and a Survey, IEEE Journal On selected Areas In Communications, Vol. 27, No. 2, pp , Feb [8] Ramjee Prasad, Fernando J. Velez, WiMAX Networks: Techno- Economic Vision and Challenges, Springer Science, [9] Chuck S., Supporting Differentiated Service Classes Queue Scheduling disciplines, White Paper, Juniper Networks, Jan [10] Mohammed Sabri Arhaif, Comparative Study of Scheduling Algorithms in WiMAX, International Journal of Scientific and Engineering Research, Volume 2, Issue2, Feb 2011 [11] Y.Subramanyam, Y.Venkateswarlu, WiMAX Base Station Scheduling algorithms, TCS White Paper, [12] Ala a Z.Al Howaide, Ahmed S. Doulat,Yaser M.Khamayseh, Performance Evaluation of Different Scheduling Algorithms In WiMAX,International Journal of Computer Science, Engineering And Applications (IJCSEA), Vol 1, No.5, Oct [13] C.Cicconetti, L.Lenzini, E.Mingozzi, C.Eklund, Quality of service support in IEEE Networks, IEEE Network, Vol. 20, No. 2, pp 50-55, March /April [14] Khaled A. Shuaib, A Performance Evaluation Study of WiMAX Using Qualnet, Proceedings of The World Congress on Engineering (WCE) July 1-3, 2009, London, U.K., pp [15] Raj Jain, D. Chiu, and W. Hawe, "A Quantitative Measure Of Fairness And Discrimination For Resource Allocation In Shared Computer Systems," DEC Research Report TR-301, Sept