WiMAX Base Station Scheduling Algorithms

WiMAX Base Station Scheduling Algorithms Worldwide Interoperability for Microwave Access (WiMAX) is one of the most important broadband wireless tech...
Author: Alvin Park
1 downloads 2 Views 3MB Size
WiMAX Base Station Scheduling Algorithms

Worldwide Interoperability for Microwave Access (WiMAX) is one of the most important broadband wireless technologies and is anticipated to be a viable alternative to traditional wired broadband techniques due to its cost efficiency. Being an emerging technology, WiMAX supports multimedia applications such as voice over IP (VoIP), voice conference and online gaming. It is necessary to provide Quality of Service (QoS) guaranteed with different characteristics. Therefore, an effective scheduling is critical for the WiMAX system. Many traffic scheduling algorithms are available for wireless networks, e.g. Round Robin, Proportional Fairness (PF) scheme and Integrated Crosslayer scheme (ICL). Among these conventional schemes, some cannot differentiate services, while some can fulfill the service differentiation with high-complexity implementation. This white paper discusses about the scheduling algorithms for Orthogonal Frequency Division Multiple Access/Time Division Multiple Access (OFDMA/TDMA) based systems, which extends the PF scheme to multiple service types with diverse QoS requirements. Keywords: CSI, QSI, OFDMA, QoS, LTL-RB, APF, 802.16e

WiMAX Base Station Scheduling Algorithms

About the Authors Subramanyam Y Subramanyam, who has a Masters degree in Avionics, has over six years of experience in design and development of Avionics software, Linux device drivers and different protocol stacks for WiMAX and LTE. He can be reached at [email protected]. Venkateswarlu Y Venkateswarlu has over two years of experience in design and development of different protocol stacks for WiMAX and LTE. He has a Masters degree in Digital Electronics and Advanced Communications. He can be reached at [email protected]

1

WiMAX Base Station Scheduling Algorithms

Table of Contents 1. Introduction

3

2. Problem Statement

3

3. Solution

4

Uplink Scheduling Algorithms

4

Information Module

5

Scheduling Database Module

6

Service Assignment Module

6

Downlink Scheduling Algorithms

7

4. Analysis

11

5. Conclusion

12

6. Acknowledgements

12

7. References

12

2

WiMAX Base Station Scheduling Algorithms

Introduction The basic IEEE 802.16 architecture consists of one Base Station (BS) and one (or more) Mobile Station (MS). BS acts as a central entity to transfer all the data from MSs in a PMP (Point to Multipoint) mode. Transmissions take place through two independent channels: Downlink Channel (from BS to MS) and Uplink Channel (from MS to BS). Uplink Channel is shared between all MSs while Downlink Channel is used only by BS. The standard defines both Time Division Duplexing (TDD) and Frequency Division Duplexing (FDD) for channel allocation. The IEEE 802.16 is connection oriented. Each packet has to be associated with a connection at MAC level. This provides a way for bandwidth request, association of Quality of Service (QoS) and other traffic parameters and data transfer related actions. The standard supports five different flow classes for QoS and the MAC supports a request-grant mechanism for data transmission in uplink direction. The standard does not define a slot allocation criterion or scheduling architecture for any type of service. It is necessary to provide a scheduling module. Five types of service flows with distinct QoS requirements [1]: l Unsolicited Grant Services (UGS): designed to support Constant Bit Rate (CBR) services like voice applications. l Real-Time Polling Services (rtPS): designed to support real-time services that generate variable size data packets

on a periodic basis, such as MPEG video, but is sensitive to delay. Real-Time Polling Services (ertPS): support real-time applications with variable data-rates, which require guaranteed data and delay, e.g. VoIP with silence suppression. l Non-Real-Time Polling Services (nrtPS): designed to support non-real-time and delay tolerant services that require variable size data grant burst types on a regular basis such as FTP. l Best Effort (BE): designed to support data streams that do not require any guarantee in QoS such as HTTP. l Extended

The standard provides specification for these different services, but does not specify any scheduling architecture. A few scheduling architectures have been reported in the literature. In this paper we discussed the different uplink and downlink scheduling algorithms.

Problem Statement Since the wireless networks are becoming an essential part of the Internet, multimedia communication applications require the network to provide quality of service for packet flows. In wired networks, Fluid Fair Queuing (FFQ) has been a popular algorithm to provide fairness among packet flows over a shared link and a number of approximation algorithms have been proposed such as WFQ, SCFQ and WF2Q. These algorithms are for wired networks and cannot be applied directly to wireless networks because a wireless channel experiences location dependent and burst channel errors. Following are some of the challenges in wireless networks: l Quality of wireless channel is typically different for different users and randomly changes with time l Wireless bandwidth is usually a scarce resource that needs to be used efficiently l Surplus amount of interference and higher error rates are typical l Mobility complicates resource allocation

3

WiMAX Base Station Scheduling Algorithms

Solution On a per flow basis, an MS (Mobile Station) requests the BS for bandwidth, for uplink. BS grants the total bandwidth for all the connections, belonging to that MS. Then the MS redistributes the sum-total of the grant among its users according to the service class of the user’s connection and its QoS requirements. This allocation scheme is known as Grant per MS (GPMS). By using Grant management Sub Header (GSH) and Bandwidth Request Header (BRH), BS controls the MS’s grant size (the amount of resources) to the size of the packet to be transmitted.

Uplink Scheduling Algorithms We are discussing more about the different uplink scheduling algorithms and describing more about the Latest Time Limit First with Reserved Bandwidth (LTL-RB). Weighted Round Robin (WRR): It is a work-conserving algorithm in which it will continue allocating bandwidth to the SSs as long as they have backlogged packets. The WRR algorithm assigns weight to each SS and the bandwidth is then allocated according to the weights. Since the bandwidth is assigned according to the weights only, the algorithm will not provide good performance in the presence of variable size packets. Earliest deadline first (EDF): It is a work conserving algorithm originally proposed for real-time applications in wide area networks. The algorithm assigns deadline to each packet and allocates bandwidth to the SS that has the packet with the earliest deadline. Deadlines can be assigned to packets of a SS based on the SS’s maximum delay requirement. The EDF algorithm is suitable for SSs belonging to the UGS and rtPS scheduling services, since SSs in this class have stringent delay requirements. Since SSs belonging to the nrtPS service do not have a delay requirement, the EDF algorithm will schedule packets from these SSs only if there are no packets from SSs of UGS or rtPS class. Weighted Fair Queuing (WFQ): It is a packet-based approximation of the Generalized Processor Sharing (GPS) algorithm. GPS is an idealized algorithm that assumes a packet can be divided into bits and each bit can be scheduled separately. The WFQ algorithm results in superior performance compared to the WRR algorithm in the presence of variable size packets. The finish time of a packet is essentially the time the packet would have finished service under the GPS algorithm. The disadvantage of the WFQ algorithm is that it will service packets even if they wouldn’t have started service under the GPS algorithm. This is because the WFQ algorithm does not consider the start time of a packet.

4

WiMAX Base Station Scheduling Algorithms

Classification and Information Module

B.Wl Request

B.Wl Request

MS M

MS1

Scheduling Database Module

Service Assignment Module

Figure 1: Uplink Scheduler Architecture The uplink (UL) scheduler comprises three modules: information module, database module and service assignment. Information Module This module performs the following functions and passes them to the scheduling data base module. l Categorizing

the packets on the MS basis l Extracting the queue size information (number of waiting packets and size of each packet of each connection from the BW-Request messages) l Deciding the arrival time: The process determines the arrival time of packets that arrived during the previous frame i.e. decides Time bound. The time bound is given by the sum of the packet’s arrival time and the packet’s maximum delay requirement (as determined by the connection QoS Parameters). The type field in the bandwidth request header indicates whether the request is incremental or aggregate. Since piggybacked Bandwidth Requests do not have a type field, PBR shall always be incremental. When the BS receives an incremental bandwidth request, it shall add the quantity of bandwidth request to its current perception of bandwidth needed for the connection. When the BS receives an aggregate bandwidth request it shall substitute its perception of bandwidth needs of the connection with the quantity of bandwidth requested.

5

WiMAX Base Station Scheduling Algorithms

Scheduling Database Module The scheduling database module serves as the information database for all the MS in the network. Service Assignment Module The service assignment module determines the uplink sub-frame allocation in terms of the number of bits per MS. The number of bits will finally be converted to the number of time slots, i.e. the units used in the information elements (IE) of the Up Link Map (UL-MAP) generation. The number of bits per time-slot is determined by the physical layer of the wireless network. The Latest Time Limit First with Reserved Bandwidth (LTL-RB) algorithm is based on an idea of reserving a minimum amount of bandwidth for each class of service during each frame-time, and then distributing the reserved bandwidth for each class of service among the related connections of that class. The surplus bandwidth is distributed among all the connections according to their instantaneous bandwidth requirements. The sharing of the surplus bandwidth among different classes follows priority logic, from highest to lowest: UGS, rtPS, ertPS, nrtPS and BE. In another word the method boils down to the following two stages: l Reserved

bandwidth of each priority class is distributed among the admitted connections of that class, according to the desired scheduling policy. l The excessed bandwidth would be allocated to those connections that have not been granted a part or total of their requested bandwidth. The surplus SB (t) comprises the left over of the reserved bandwidth of each traffic class, if any, and also the remainder of the bandwidth (P (t)). The main purpose of the second round of scheduling is to make best use of the system utilization and to supply soft QoS, by allocating the left over bandwidth of one class of service to another class of service, which requires more bandwidth than its reserved bandwidth, due to the arrival of traffic in bursts. The computation of the amount of reserved bandwidth to be allocated to each class reacts the guiding principle of the scheduler. Reserving a zero amount of bandwidth for each class of service would turn the scheduler into a purely strict priority scheduler and thus reserving a huge quantity of bandwidth for lower priority classes would have a negative impact on the QoS support of higher priority classes. The values of these reserved bandwidths can be adjusted dynamically. The allocation of bandwidth to the UGS connections in our scheduling scheme requires the allocation of fixed size data grants at periodic intervals [2]. The scheduling methods, which are considered here for allocation of bandwidth within the connections of each class of service, are Latest Time Limit First with Reserved Bandwidth (LTL-RB). The LTL-RB algorithm is carried out in four stages.

6

WiMAX Base Station Scheduling Algorithms

During the initial stage the scheduler inspects each non-empty (MS1 -- MSM) queue and sorts the BW requests in the ascending order of their Expiry Time (ET) and services (rtPS, ertPS, nrtPS, be). During the next stage the requested data size of the first BW request packet in the rtPS queue is checked. If it is less than or equal to rtPS, the reserved bandwidth is reduced by the size of the BW request, and all the connection's requested bandwidth are allocated to it. Otherwise, the BW-request is reduced by the size of rtPS and the connection's bandwidth requirement is somewhat allocated to it. This process is repeated until either the rtPS is no longer greater than zero or the request queue is empty. In the case that the request queue is empty but rtPS is greater than zero, rtPS will be added to the sum of leftover (bit) bandwidth at time t (l (t)). When any of the above condition occurs, the scheduler goes to service the ertPS request queue. Once the ertPS connections being serviced the scheduler goes to service nrtPS and then to the BE request queue. During the third stage, the scheduler allocates the surplus bandwidth SB (t) to the connections, which have not been serviced in the first round of bandwidth allocation, starting from the highest priority queue. The scheduler moves to service the next priority queue if SB (t) is greater than zero and if the queue is not empty. The procedure shown in algorithm continues for the third stage of scheduling, but this time instead of distributing the reserved bandwidth, the SB (t) would be distributed within the connections that still require bandwidth. During the fourth stage, when the bandwidth size to be allocated to each connection is determined, the allocations in the frame are made in such a way that each MS gets adjacent allocation time. This is because in IEEE 802.16e the bandwidth allocation is per MS and not per connection. Among the connections with the same ET and within the same service class, bandwidth can be granted as per the following strategies. The connection with higher queue size is chosen because this connection experiences higher levels of congestion, and its buffer management module is more likely to drop its packets due to overflow.

Downlink Scheduling Algorithms Downlink (DL) Scheduler in BS distributes the entire downlink bandwidth among downlink connections. BS uses Adaptive Proportional Fair (APF) scheduling scheme to schedule the control messages and data packets. When the uplink sub-frame ends, BS first broadcasts DL-MAP and UL-MAP, then the RANG-RSP messages, then the REGREP messages, then the CONN-RSP messages and then starts sending downlink data packets to MSs. The downlink channel is an always broadcast channel. BS Downstream Generator sends specific amounts of data from each downlink connection according to the output of the Downlink Scheduler. This module is also responsible for sending messages generated in the Uplink Scheduler. We are discussing the different downlink scheduling algorithms and describing more about the Adaptive Proportional Fairness (APF) scheduling algorithm.

7

WiMAX Base Station Scheduling Algorithms

Round Robin (RR): It is one of the simplest scheduling algorithms designed especially for a time sharing system, where the scheduler assigns time slots to each queue in equal portions without priority. Once a queue is served, it is not visited again until all the other MSs in the system have been served. RR offers same visit times to all queues regardless of their channel conditions, resulting in the fairness and low channel capacity. Another disadvantage of RR scheme is no multi-user diversity gain. Also, it cannot guarantee different QoS requirements for each queue. Proportional Fairness (PF): The essential goals of this packet scheduling scheme are to enhance the system throughput as well as provide fairness among the queues under consideration. Although PF is simple and efficient, it cannot guarantee any QoS requirement such as delay and delay jitter due to its original design for saturated queues with non real-time data service Integrated Cross-layer Scheduling: Both RR and PF cannot manage the resource allocation and grants an appropriate QoS per connection. The scheduler bases on a priority function for each queue, where the priority metric of each queue is updated according to its service status and channel condition in the PHY layer. Thus, the scheduler can provide the prescribed diverse QoS guarantees. This scheduler is hard to be practically deployed due to its high implementation complexity and the delay performance of each type is not good. The Cross-Layer Scheduling Architecture is shown in Figure 2. The scheduling algorithm at the MAC layer is modeled as an optimization problem with respect to some physical layer constraints and application QoS constraints. At every timeslot, the scheduling algorithm has to produce rate allocation r = ( r1 ,...., rK ) and power p = ( p 1 ,....., p K ) for all the k users, which is based on the observation of the current channel state information (CSI) from the physical layer and the queue state information (QSI) from the application layer. Rate allocation and power allocation are selected so that they optimize some system objectives, summarized below. Application Layer Queues

MAC Layer (Cross Layer Optimization)

Scheduling output

Channel State Information

PHY Layer

Figure 2: Cross-Layer Scheduling Architecture

8

WiMAX Base Station Scheduling Algorithms

A multi-user wireless network can be modeled by a dynamic system and the states evolution given by the following:

~ State Information (CSI): The channel fading states H (t ) = [ H1 (t ),....., H K (t )] where H k (t ) is the th channel fading process between the base station and the k user. l Queue State Information (QSI): The queue length is Q( t) = [ Q1 (t ),....., QK (t )] where Qk (t ) denotes the number of untransmitted packets in the buffer of the user k at time t. l Channel

The Scheduler structure of downlink is depicted in Figure 3. Adaptive Proportional Fairness (APF) scheduling scheme, is introduced, which aims at extending the PF [3] scheduling to the real time service and provides various QoS requirements. The scheduling scheme is based on the Grant Per Type-of-Service (GPTS) principle, which aims at differentiating the delay performance of each queue. A novel priority function is devised for all the QoS guarantee queues, including UGS, rtPS, nrtPS and ertPS, for allocating time slots on the queues with the highest priority value. At the time interval t, the priority function for queue i is defined as:

Packet Classifier USER MS 1 Traffic Types

MS K

UGS rtPS nrtPS BE ertPS

UGS rtPS nrtPS BE ertPS MS K

Control Messages MS1 MS K

Broadcast Messages MS1

Downlink Scheduler

MS 1

MS 2

MS 3

MS K

Figure 3: Downlink Packet Scheduler r ( t) m (t ) = i i Ri ( t) / C i (t ) M i (t )

(1)

9

WiMAX Base Station Scheduling Algorithms

Where Mi is the minimum rate requirement, Ci (t) is the number of connections of the ith queue and ri (t) is the transmission capacity at time t, which is determined by the channel quality. Each queue corresponds to one QoS requirement class, respectively. The queue having the highest value of µi(t) is scheduled. The estimated average throughput of scheduled queue i (denoted as Ri (t)) is updated by the following simple exponential smoothing model.

1 1 R i (t + 1) = (1 ) Ri (t ) + ri (t , n ) ( 2) å Ti T i nÎ W k (t ) Where ri(t) is the supportable data rate for user i in sub carrier n at time slot t and Ri(t) is the average throughput th received by the i user up to slot t. W i (t ) indicate the set of sub-carriers in which user i is scheduled for transmission at time slot t. Ti is a pre-defined system parameter corresponding to each class. Rather than making Ti constant in the PF scheme, we assign different values of Ti to each type to execute the different applications. We highlighted on explaining the preference metric available to support diverse services with different QoS requirements. We assume the variation of channel conditions can allow us to select the same Ti for the same service type. The MS that are close to the BS, will not sacrifice the transmission of the others. The key parameter of this algorithm is Ti in Eq. (2). The role of Ti is to distinguish the priority for different types of services. With a small Ti , the preference metric [4-5] of queue i fluctuates significantly, so that queue i is visited frequently. This feature is critical to distinct the queue with the delay constraint. For example, in the case Ti =2, according to Eq. (2), if the queue has not been served by the scheduler at previous n time frames, decreases dramatically, which lead queue i to win the transmission opportunity at the current time frame more likely. In contrary, with a large Ti , (for instance, when Ti =100) Ri(t) decreases slightly although the queue has not been visited for a few time frames. Thus, the channel quality ri(t) is the main determining factor for the priority of each queue. In this case, the system throughput is enhanced and efficient bandwidth utilization is achieved. However, the scheme is not susceptible to the latency of the connections. With such a design, the task is finding a proper Ti to achieve the balance of delay and throughput for each class according to QoS requirements. th

Quantity RI(t) / Ci(t) specified in Eq. (1) normalizes the throughput of each connection in the i queue. Ri(t ) / (Ci (t)Mi) is an indicator of data rate satisfaction and also reflects the delay satisfaction for real time connections. For real-time service, Ri (t) evaluated with a small Ti is sensitive to the waiting time of the data in queue i. Large value of Ri (t) indicates high degree of delay satisfaction, which leads to low priority. Therefore, a variety of QoS demands for real-time and non-real-time applications are unified to Ri(t) / (Ci(t)Mi), which plays an essential role in reflecting the instantaneous bandwidth requirements of queue i.

10

WiMAX Base Station Scheduling Algorithms

Analysis The algorithm LTL-RB in uplink follows priority rule to meet delay and loss requirements of different classes. It also tries to maintain its fairness by reserving a minimum amount of bandwidth for each class of service, during each frame time. The size of this amount of reserved bandwidth for each class of service reacts with a degree of scheduler fairness towards that class. It can also be said that the size of the reserved bandwidth is a trade-off between providing an improved QoS support for higher priority classes and being fair to lower priority classes as compared with the other uplink algorithms discussed in this paper. Downlink algorithm together considers the effect of the current channel conditions and the transmission fulfilment of the previous time frames. Compared with RR, PF and ICL schemes, APF outperforms in service differentiation and QoS provisioning by choosing an appropriate set of Ti. APF is flexible to the system size in terms of the number of accommodated MSs. Comparison of different scheduling algorithms for both uplink and downlink are shown in the below tables. Comparison of uplink scheduling algorithms in WiMAX

Fairness Scheme

Average Throughput

Average Delay

Intra class

Inter class

WRR

Medium

Medium

High

High

EDF

High

Low

High

Low

WFQ

Medium

Low

Low

Low medium

LTL- RB

High

Low

High

High

Comparison of downlink scheduling algorithms in WiMAX Scheme

Average Throughput

Average Delay

Fairness

Frame Utilization

RR

Low

High

Low

Low

PF

High

Low

Medium

High

ICL

Medium

Medium

Low

Medium

APF

High

Low

High

High

11

WiMAX Base Station Scheduling Algorithms

Conclusion Latest Time Limit First with Reserved Bandwidth (LTL-RB) algorithm for uplink and Adaptive Proportional Fairness (APF) scheduling algorithm for downlink having the advantages over the remaining algorithms discussed in this paper. So we implemented these algorithms in TCS for our WiMAX solution.

Acknowledgements This work has been greatly supported by Dr.Girish Chandra and Praveen (R&D Team) from Tata Consultancy Services.

References 1. IEEE 802.16e Broadband Wireless Access systems 2. Quality of Service Support in IEEE 802.16 Broadband Wireless Access Networks By Ehsan Asadzadeh Aghdaee 3. Xiaojing Meng "An Efficient Scheduling for Diverse QoS Requirements in WiMAX" A thesis presented to the University of Waterloo. Waterloo, Ontario, Canada, 2007. 4. P.Viswanath, D.N.C.Tse, and R. Laroia, “Opportunistic beamforming using dumb antennas,” IEEE Transactions on Information Theory, vol. 48, no.6, pp. 1277-1294, Jun. 2002. 5. J.M. Holtzman, “Asymptotic analysis of proportional fair algorithm,” in Proc. IEEE PIMRC 2001, San Diego, CA, pp. 33-37. 6. A Performance Study of Uplink Scheduling Algorithms in Point to Multipoint WiMAX Networks by Pratik Dhrona. 7. An Efficient Scheduling For Diverse QoS Requirements in WiMAX by Xiaojing Meng 8. Quality of Service Support in IEEE 802.16 Broadband Wireless Access Networks by Ehsan Asadzadeh Aghdaee

12

About HiTech Industry Solution Unit

About Tata Consultancy Services (TCS)

TCS’ HiTech Industry Solution Unit comprises of Semiconductors,

Tata Consultancy Services is an IT services, business solutions and

Electronics, Computer Platforms & Services, Software industry and

outsourcing organization that delivers real results to global

Professional Services. At TCS, we leverage our experience in

businesses, ensuring a level of certainty no other firm can match.

Engineering Services, Business Process Transformation, end-to-end

TCS offers a consulting-led, integrated portfolio of IT and IT-enabled

IT Solutions and Infrastructure Services to provide comprehensive

services delivered through its unique Global Network Delivery

solutions that will help the High Tech firms and manufactures

ModelTM, recognized as the benchmark of excellence in software

accelerate product innovation, achieve operational excellence,

development.

attain greater profitability and maintain market leadership. Our proven consulting capabilities, Extensive engineering expertise, and in-house innovation labs provide breakthrough transformation of product and service portfolios. Our recent investments include dedicated labs and infrastructure in support for convergence solutions, embedded printer solutions, storage optimization and High Tech Center of Excellence based in

A part of the Tata Group, India’s largest industrial conglomerate, TCS has over 143,000 of the world's best trained IT consultants in 42 countries. The company generated consolidated revenues of US $6 billion for fiscal year ended 31 March 2009 and is listed on the National Stock Exchange and Bombay Stock Exchange in India. For more information, visit us at www.tcs.com.

Eindhoven (The Netherlands).

TCS Design Services M 0709

E-mail: [email protected]

Subscribe to TCS White Papers TCS.com RSS: http://www.tcs.com/rss_feeds/Pages/feed.aspx?f=w Feedburner: http://feeds2.feedburner.com/tcswhitepapers

All content / information present here is the exclusive property of Tata Consultancy Services Limited (TCS). The content / information contained here is correct at the time of publishing. No material from here may be copied, modified, reproduced, republished, uploaded, transmitted, posted or distributed in any form without prior written permission from TCS. Unauthorized use of the content / information appearing here may violate copyright, trademark and other applicable laws, and could result in criminal or civil penalties. Copyright © 2009 Tata Consultancy Services Limited

www.tcs.com

Suggest Documents