Scheduling Algorithms for Super 3G Jean-Christophe Laneri Kungliga Tekniska Högskolan Radio Communications Laboratory Master Thesis Project Advisor: Hannes Ekström Examiner: Slimane Ben Slimane (Ericsson Research) (KTH) Master Thesis Presentation 1
Agenda Scheduling Algorithms for Super 3G The Big Picture Algorithms Simulation Results Conclusions Master Thesis Presentation 2
Super 3G Initial deployment in the 2009-2010 time perspective Evolution of the 3GPP Radio Access Network (RAN) Traffic carried on to of IP: Enhancements for packet-based services Targets related to this work High Peak & cell-edge data rates Spectrum Efficiency improvement & Spectrum Flexibility Improved Service Provisioning Possible ways of enforcing these objectives which concern this project Reduced number of Network Nodes Differentiation of Services over a Shared Infrastructure Physical Layer Downlink: OFDM with frequency adaptation Master Thesis Presentation 3
Motivations and Problem Definition Providing Quality of Service over Shared Channels Schedulers are used in order to Divide the Resources between the users over Shared Channels Provide Spectrum Efficiency Fulfill Service Requirements We study Downlink Scheduling Algorithms within the following context Multi-user OFDMA environment (with frequency adaptation) Services Differentiation Realistic Traffic Models We aim at Verifying if the QoS policy profiles can be enforced Evaluating some scheduling algorithms characterized by different level of fairness (in terms of user data rates) Investigating the tradeoffs between network capacity and user fairness Master Thesis Presentation 4
QoS Concept Realization of DiffServ for 3GPP access networks Principle Mark each packet at the network edge with a a Flow-Class Identifier (FC-ID) FC-IDs permit to identify the QoS class of packets at each network node Associate a Policy Profile to each FC-ID FC-IDs are grouped into three categories SIGnaling Guaranteed Bit Rate: Policy Profile = [strict priority] Best Effort: Policy Profile = [Committed Rate, Priority] Aggregate Cell Throughput Internet Access Corporate Access CR= 30% CR= 70% CR= 30% CR= 70% Video (Pr=2) Allocated to Signaling VoIP (Pr=1) Time Master Thesis Presentation 5
Scheduling Architecture Dividing the bandwidth between the FC-IDs BE Inter-FC-ID Scheduler GBR Inter-FC-ID Scheduler S3G Scheduler Master Thesis Presentation 6
Scheduling Framework Radio Resources, Algorithm Inputs and Design Parameters AR-process with 1 s memory Perceived User Throughputs Channel Estimate for all OFDM subbands Available at every tti Scheduler 20 MHz bandwidth 1 81 Allocation Algorithm FC-ID Policy Profiles GBR <FC-ID><Priority><delay cst> BE <FC-ID><CR><Priority> Link Adaptation Estimate the transferable amount of bits given a scheduling allocation Scheduling Decision <user><fc-id><number of Bits> Master Thesis Presentation 7
Resource Allocation Algorithms (1/2) Dividing the bandwidth between the users: Intra-FC-ID Maximum Signal to Interference Ratio (Max SIR) Proportional Fair (PF) Exponential Rule (ER) Modified Exponential Rule (ER 2 ) VoIP Scheduler (VoIP) Master Thesis Presentation 8
Resource Allocation Algorithms (2/2) Dividing the bandwidth between the users: Intra-FC-ID Fair Throughput Smaller throughput average Larger throughput User 7 User 5 User 2 User 6 User 9 User 3 While users in U1 = {7,5,2} have data to receive Iterate over U1, giving the best available chunk While users in U2 = {6,9,3} have data to receive Iterate over U2, giving the best available chunk Master Thesis Presentation 9
Scenarios & Performance Measures Algorithms Validation and Realistic Traffic Evaluation Scenarios Fully Loaded System (Infinite source only PHY/MAC is simulated) File Transfer (10 MB fixed file size, PHY/MAC/RLC/IP/TCP simulated) Web-Browsing (500 kb fixed file size, PHY/MAC/RLC/IP/TCP simulated) Voice Over IP (32 B frame every 20 ms, PHY/MAC/RLC/IP/UDP/RTP simulated) Performance Measures Cell Throughput (Information bits) Link Utilization (Percentage of used chunk) User Throughput Fairness (Jain fairness index) Mean User Throughput VoIP capacity: 90 % of the users with a delay below 50 ms Master Thesis Presentation 10
Simulation Results QoS Concept Verification for the Best Effort Policies GBR service is emulated 15.55 % BE traffic gets a varying CR=30 % percentage of the resources. Two BE FC-IDs, with Committed Rates of 70 % and 30 % BE Policies are enforced! Percentage of Allocated Resources CR=70 % Emulated GBR 36.3 % 48.15 % Time (s) Master Thesis Presentation 11
Simulation Results Full Buffer Scenario (PHY/MAC only-100 users 400 s) Traffic belonging to one BE FC-ID Is simulated. Tradeoff between Cell and User Throughputs Fair Throughput Fairness = 0.899 Maximum SIR Fairness = 0.121 Proportional Fair Fairness = 0.435 ER 2 Fairness = 0.245 ER Fairness = 0.418 1Mbps 10Mbps 100Mbps Master Thesis Presentation 12
Simulation Results File Transfer (PHY/MAC/RLC/IP/TCP-10 MB files-20 min) Cell Fairness Throughput Mean Link Utilization Throughput Arrival Rate (users/s) Fair Throughput Maximum SIR Proportional Fair Modified ER Arrival Rate (users/s) Master Thesis Presentation 13
Simulation Results Web Browsing (PHY/MAC/RLC/IP/TCP-500 kb pages-20 min) Web-Effect: users with low serving time (high SIR) monopolize the system Traffic with interactive characteristic Conclusion: A fair scheduler could be profitable Master Thesis Presentation 14
Simulation Results VoIP (PHY/MAC/RLC/IP/UDP-32 B frames-20 min) Delay Constraint: 90% of the received packets with a delay under 50 ms 90 th percentile of the packet delays 50 ms delay tolerance Fair Throughput Max SIR VoIP Number of Users Master Thesis Presentation 15
Conclusions and Future Works QoS Concept Validated Service Specific Scheduler Conclusions Policy Profiles can be enforced Proposed schedulers validated with the Full-Buffer Scenario Impact of realistic traffic: service-dependent scheduling algorithms Radio oriented method for background downloads Fair approach for interactive scenarios Delay aware scheduler for conversational services QoS concept well suited for this type of differentiation Future Works Controlling the scheduling decision as a function of what the users perceive (fairness, mean throughput) and load. Resources Allocation for more than one tti Master Thesis Presentation 16
Master Thesis Presentation 17