PLANETE group, INRIA Sophia-Antipolis July 1, 2003 Adaptive Channel allocation for QoS Enhancement in IEEE Wireless LANs Presented by: Mohammad Malli Advisors: Qiang Ni, Thierry Turletti, and Chadi Barakat Université de Nice Sophia-Antipolis Ecole Doctorale STIC DEA Réseaux et Systèmes Distribués
July 1, Outline ò IEEE and e ò Problems and Solutions ò Simulation Topologies and Parameters ò Our scheme: Adaptive EDCF ò Our scheme: Adaptive DCF ò Conclusions and Future Work
July 1, IEEE New technology ò Provide end-users the benefits of increased mobility & productivity ò Enable network connectivity at locations where cabling is either difficult or costly to install IEEE Logical Link Control (LLC) IEEE Media Access Control (MAC) IEEE Physical Layer (FHSS, DSSS, IR)
July 1, IEEE MAC two MAC layer uses two kinds of protocols to access to the medium ò DCF : Distributed Coordination Function is used to support asynchronous data transmissions ò PCF : Point Coordination Function is designed for time-bounded multimedia applications
July 1, CSMA/CA Source Destination Data Others Ack Data SIFS DIFS Backoff Time Defer Access Time Backoff_Time = Rand(0, CW) * aSlotTime Backoff Counter must be decreased each slot time by one slot time whenever the channel is idle å CW is initially set to CWmin å CW is doubled after a failed transmission å CW is set to Cwmin, when the packet is successfully transmitted
July 1, DCF Limitations service differentiation DCF is unsuitable for real time applications because it doesn’t support service differentiation collision rateidle slots DCF suffers from significant throughput degradation and high delay at high load due to high collision rate and wasted number of idle slots in each backoff contention cycle Collisions Idle backoff slots Idle backoff slots (at each contention period) Virtual transmission time SIFSACKDIFS ACKSIFS
July 1, FCR (Fast Collision Resolution) ò Proposed by Florida U. (Infocom ’03) no service differentiation ò Extends the basic DCF (no service differentiation), to improve the throughput ò Main features: Static Backoff Threshold - Static Backoff Threshold value = 2 * (CWmin + 1) - 1 Increasing CW - Increasing CW when the channel is busy during deferring periods ò Weakness: å Backoff Threshold must be adapted to the medium state because during high load, the period of exponential state must be shorter to reduce aggressivity å When the channel is busy, it is better that the node waits with its remaining backoff time because doubling the CW during deferring periods increases the number of idle slots in low and medium load cases
July 1, IEEE e Upcoming IEEE e MAC: ò HCF: Hybrid Coordination Function has Controlled Channel Access Mechanism, it is used in infrastructure network EDCF ò EDCF: Enhanced Distributed Coordination Function doesn’t need a central coordinator point, it is used in Ad-hoc network service differentiation Extends the basic DCF to support service differentiation
July 1, EDCF - Enhanced DCF DCF EDCF CW CW[TCi ] AIFS[TCi] Transmission attempt Many classes provide per flow differentiation Class0 Class2 Class1 Class7 TC7 TC2TC1 TC0 DIFS Internal Scheduler (Resolve Virtual Collisions)
July 1, EDCF Limitations EDCF fails to provide QoS at high load Bad Video QualityLow Total Throughput AudioVideo Background Throughput (in B/s) Time (in sec)
July 1, Our approach: Adaptive EDCF (1) QoS ò improve the QoS for multimedia applications in all medium states total throughput increase the total throughput in all medium states Goals : Idea : protect Audio and Video To protect Audio and Video transmissions, best effort queue increases its CW larger and reset a new backoff time, when it senses the channel is busy, during deferring periods decrease the wasted number of idle slots To decrease the wasted number of idle slots due to backoff in each contention cycle, a queue must decrease faster its backoff time after it senses the channel idle during a certain time í Each extension will help to realize the above goals
July 1, AEDCF(2): 1st extension Our Solution : First, extend the Fast Backoff mechanism, proposed for DCF in Florida U., to an adapted approach that differentiates between the different priority levels 2 Backoff states : Linear å Linear decrease (old) Exponential å Exponential decrease (new) - Adapted to medium state - Differentiate between traffic classes Linear stateexponential state 0 Backoff_ThresholdBackoff_Time Backoff_Counter decrease Slot T.
July 1, AEDCF (3): 2nd extension Second, increase the contention window size and reset a new backoff time, when the channel is sensed busy, during deferring periods : CW[pri] = min(CWmax[pri], 2 * CW[pri]) Priority CWmax í Low priority flows will be punished High priority flows will be protected í High priority flows will be protected
July 1, Simulations Topology and parameters (1) Node0 Node 1 Node2 Node n Audio Video Background Audio Video Background Low load : n = 5, 1 Mbytes/s Medium load : n = 11, 2.5 Mbytes/s High load : n = 15, 3.5 Mbytes/s Medium Bandwidth = 4.5 Mbytes/s
July 1, Simulation Topologies and parameters (2) MAC parameters for the three flows
July 1, AEDCF: Flows Throughput Good multimedia flow performance in medium and high load cases Also, Background flows have better throughput than in EDCF case Throughput with AEDCF in 11 nodes topology Throughput with AEDCF in 15 nodes topology Throughput (in B/s) Time (in sec)
July 1, AEDCF: Total Throughput highest Total Throughput Our scheme also provides highest Total Throughput in high load case Total Throughput in 15 nodes topology With our AEDCF scheme, the T.T is higher about 55 % more than with EDCF and 10 % more than DCF in this high load topology 3.5 Mbytes/s Total sending rate Total Throughput (in B/s) Time (in sec)
July 1, Our approach: Adaptive DCF (1) adapted fast backoff Extends DCF by our adapted fast backoff approach Throughput with DCF Throughput with FCR Throughput with ADCF best medium utilisation ADCF provides best medium utilisation in this medium load case Throughput (in B/s) Time (in sec)
July 1, ADCF (2): High load service differentiation We still need service differentiation to maintain a stable multimedia flows quality in high load Throughput with DCF Throughput with FCR Throughput with FCR Throughput with ADCF Throughput with ADCF Throughput (in B/s) Time (in sec)
July 1, ADCF (3): Total Throughput highest Total Throughput In medium and high load cases, ADCF provides the highest Total Throughput Total Throughput in 11 nodes topology Total Throughput in 11 nodes topology Total Throughput in 15 nodes topology Total Throughput in 15 nodes topology In this medium load case, DCF is better than FCR FCR is better than DCF only in high load case 3.5 Mbytes/s Total sending rate 2.75 Mbytes/s Total sending rate Total Throughput (in B/s) Time (in sec)
July 1, Conclusions and Future Work ò QoS support in IEEE and e WLANs is not good enough Adaptive EDCF ò We propose an extension to the proposed e EDCF: Adaptive EDCF å Uses adaptive fast backoff mechanism å Provides more transmission opportunity to multimedia applications and higher total throughput during high load situations Adaptive DCF ò we propose an extension to the standard DCF: Adaptive DCF å Uses adaptive fast backoff mechanism å Provides better medium utilisation and higher total throughput in medium and high load cases å It is not good enough for multimedia applications in high load state î In this case, it is better to use AEDCF ò Future work: Analytic modeling & Real Experimentation
July 1, Thank you Q & A