doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 1 EDCA Parameters Selection to Optimally Provide QoS in IEEE s Mesh WLANs Notice: This document has been prepared to assist IEEE It is offered as a basis for discussion and is not binding on the contributing individual(s) or organization(s). The material in this document is subject to change in form and content after further study. The contributor(s) reserve(s) the right to add, amend or withdraw material contained herein. Release: The contributor grants a free, irrevocable license to the IEEE to incorporate material contained in this contribution, and any modifications thereof, in the creation of an IEEE Standards publication; to copyright in the IEEE’s name any IEEE Standards publication even though it may include portions of this contribution; and at the IEEE’s sole discretion to permit others to reproduce in whole or in part the resulting IEEE Standards publication. The contributor also acknowledges and accepts that this contribution may be made public by IEEE Patent Policy and Procedures: The contributor is familiar with the IEEE 802 Patent Policy and Procedures, including the statement "IEEE standards may include the known use of patent(s), including patent applications, provided the IEEE receives assurance from the patent holder or applicant with respect to patents essential for compliance with both mandatory and optional portions of the standard." Early disclosure to the Working Group of patent information that might be relevant to the standard is essential to reduce the possibility for delays in the development process and increase the likelihood that the draft publication will be approved for publication. Please notify the Chair as early as possible, in written or electronic form, if patented technology (or technology under patent application) might be incorporated into a draft standard being developed within the IEEE Working Group. If you have questions, contact the IEEE Patent Committee Administrator at. Date: Authors:

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 2 Outline Challenges and Overview of IEEE s Mesh WLAN Overview of EDCA Analytical Model of EDCA Simulation Setup Observations How to Handle Each Observation? Experimental Setup Conclusion

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 3 Outline Challenges and Overview of IEEE s Mesh WLAN Overview of EDCA Analytical Model of EDCA Simulation Setup and Results Observations How to Handle Each Observation? Experimental Setup and Results Conclusion

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 4 Challenges in IEEE s Mesh Networks Challenges –Lack of central coordinators – Each node is a coordinator –Multi-hop environments For Absolute QoS Guarantees –Need to determine the optimal route and perform a hop-by-hop reservation For Prioritized QoS –Need to maximize throughput and maintain pre-specified throughput ratios between different flows or classes, need to compute optimal CW for each class –Optimal CW j depends (1 or multi)-hop neighboring information: such as number of nodes in each class and their traffic characteristics

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 5 MAC layer issues in s Mesh Networks EDCA will be the basis of IEEE s MAC 1 –EDCA will be used to provide prioritized QoS –Nodes (say i,j) forwarding their traffic to a particular node, (say x) will have to have receive the EDCA parameters from Node x –If the forwarding node (x) is allowed to set parameters to individual flows then it can provide absolute QoS for a particular flow Enable congestion control 1 –Simple hop by hop congestion control enabled at each Mesh Point (MP) –Need to curtail a flow if it has violated its QoS agreements –Because the forwarding node has reduced channel capacity, it may ask all nodes that forward traffic to it to reduce their rates. (By form of signaling newer EDCA parameters) 1 DCN: IEEE /0329r0: Joint SEE-Mesh/Wi-Mesh Proposal to TGs Overview

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 6 Example WLAN Mesh Assume a simple clustering structure –Mesh Portals (MPs) are at level-0 –Level-i MP nodes coordinate level-(i+1) MP nodes A level-i MP node estimates the number of nodes in each class by communicating with its parent and sister nodes, and by monitoring its children nodes The level-i MP node computes the optimal CW/AIFS for each traffic class and broadcasts them to its children nodes Level-(i+1) MP nodes use the designated CW/AIFS to access the channel MP 1 MP 2 MP 21 MP2 12 MP 211 MP 212 Now let us understand the performance of IEEE e EDCA ……

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 7 Outline Challenges and Overview of IEEE s Mesh WLAN Overview of EDCA Analytical Model of EDCA Simulation Setup and Results Observations How to Handle Each Observation? Experimental Setup and Results Conclusion

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 8 Overview of EDCA: Enhanced Distributed Control Access DCF enhanced with QoS Figure 1. Four access categories with different QoS parameters Table 1. Default EDCA Parameters aCW min = 32, aCW max = AIFS = SIFS + AIFSN*aSlotTime

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 9 Key parameters in EDCA Contention window size (CW) Arbitrary inter-frame space (AIFS) Transmission opportunity limit (TXOP) Decide the probability of gaining the channel access Decide the time of occupying the channel

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 10 Outline Challenges and Overview of IEEE s Mesh WLAN Overview of EDCA Analytical Model of EDCA Simulation Setup and Results Observations How to Handle Each Observation? Experimental Setup and Results Conclusion

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 11 The system in view of the channel Contention zone j: –Consecutive slots in which only the first j classes are eligible to access the channel –Numbering the slot from the first slot after a busy period + SIFS, the j-th contention zone starts from the a j -th idle slot and ends at the (a j+1 -1)-th idle slot. Figure 3. Following every busy period, the slots can be divided into contention zones. a j = AIFSN j + 1. a 1 = 3, a 2 = 4, a 3 = 8

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 12 Describe the channel state transition Assume attempt probability of each class,  j, are known at this step. The stable probabilities can be readily derived. Details (see the Infocom2006 paper at Success states Collision states Idle states Figure 4. The discrete Markov chain that describes the channel state transition

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 13 Other elements in the model Assumptions: –Saturation condition: every node is back-logged –Every node has the same view of the channel state An iterative algorithm to obtain average contention window size, and attempt probabilities  j –Given [CW MIN (j), CW MAX (j)] and L j Derive the performance, e.g., the throughput –Expected slot length –Throughput of each class Packet payload Success probability of class j

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 14 Outline Challenges and Overview of IEEE s Mesh WLAN Overview of EDCA Analytical Model of EDCA Simulation Setup and Results Observations How to Handle Each Observation? Experimental Setup and Results Conclusion

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 15 Simulation Setup Tools: –Simulator: ns-2, extended with EDCA –Analytical results: use Matlab PHY/MAC parameters: –Data rate 200 Mbps (all throughput results are normalized to it) –Slot time: 8  sec –SIFS: 10  sec –Retry times limit: 7 Traffic: –All classes have the same number of nodes –All nodes transmit packets to a sink node –CBR (Constant-Bit-Rate), rate large enough to backlog every node –Packet size is 1024 bytes

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 16 Impact of contention window size, CW Figure 5-1. Three classes with the same AIFS: CW 1, 2, 3 = [8, 16], [16, 32], [32, 64] The temporary increase (will eventually decrease) is caused by the non-uniform access to the post-busy slot.

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 17 Impact of AIFS Figure 5-2. Two classes with the same CW: AIFS 1, 2 = 2, 3

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 18 Impact of CW and AIFS combined Figure 5-3. Four classes: CW 1, 2, 3, 4 = [8, 16], [16, 32], [32, 1024], [32, 1024], AIFS 1, 2, 3, 4 = 2, 2, 3, 7. (Default EDCA parameters)

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 19 Outline Challenges and Overview of IEEE s Mesh WLAN Overview of EDCA Analytical Model of EDCA Simulation Setup and Results Observations How to Handle Each Observation? Experimental Setup and Results Conclusion

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 20 Observation 1 Observation 1 Higher priority traffic with smaller AIFS value can easily grab most of the bandwidth and starve other traffic. Higher priority traffic with smaller AIFS value can easily grab most of the bandwidth and starve other traffic. Figure 6-1. Two classes with the same CW but different AIFS. Left: AIFS 1 = 2, AIFS 2 = 3. Right: AIFS 1 = 2, AIFS 2 = 5.

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 21 Figure 6-2. Three classes: CW 1, 2, 3 = [8, 16], [16, 32], [32, 64]. Left: throughputs. Right: throughput ratios of class 2/1, and class 3/1. Observation 2 Observation 2 Bandwidth allocation fails to stay stable – it varies as the system load (the number of nodes) varies. Bandwidth allocation fails to stay stable – it varies as the system load (the number of nodes) varies. ratio

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 22 Observation 3 Observation 3 Bandwidth is under-utilized – the system is not operating at optimal condition (a problem known in DCF). Bandwidth is under-utilized – the system is not operating at optimal condition (a problem known in DCF). Figure 6-3. Existence of optimal operating points that can maximize the throughput. Optimal operating points

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 23 Outline Challenges and Overview of IEEE s Mesh WLAN Overview of EDCA Analytical Model of EDCA Simulation Setup and Results Observations How to Handle Each Observation? Experimental Setup and Results Conclusion

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 24 Suggestion: – Small AIFS has to be carefully used to avoid burst contention –Use by real-time traffic only for admission and reservation –Real-time traffic, once admitted and made a reservation, access the channel using reservation-based access (e.g. poll-based) –Normal contention-based access use the same AIFS Observation 1 Observation 1 Higher priority traffic with smaller AIFS value can easily grab most of the bandwidth and starve other traffic. Higher priority traffic with smaller AIFS value can easily grab most of the bandwidth and starve other traffic.

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 25 Leverage our theoretical model to achieve –Deterministic weighted bandwidth allocation, and –Maximize the bandwidth utilization Observation 2 Observation 2 Bandwidth allocation fails to stay stable – it varies as the system load (the number of nodes) varies. Bandwidth allocation fails to stay stable – it varies as the system load (the number of nodes) varies. Observation 3 Observation 3 Bandwidth is under-utilized – the system is not operating at optimal condition (a problem known in DCF). Bandwidth is under-utilized – the system is not operating at optimal condition (a problem known in DCF).

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 26 Theoretically suggested CWs Theorem 1: For M classes traffic with the same AIFS value, to achieve proportional bandwidth allocation: r j, which is defined as the ratio of the throughput of class j and that of class 1, and maximum bandwidth utilization, the CW of each class shall be set as follows: where and T D ’ is the duration of a successful transmission in the unit of slots. r 1  1.

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 27 Numerical and simulation results Figure 8. The throughput ratio among different traffic classes before (left) and after (right) optimization based on the theoretical model

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 28 Simulation results Figure 10. The total throughput and the throughput attained by each class in the presence of two real-time streams and two classes of best-effort traffic. Left: pre-set bandwidth allocation ratio r 3/2 = 0.5 Right: pre-set bandwidth allocation ratio r 3/2 = 0.25

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 29 Outline Challenges and Overview of IEEE s Mesh WLAN Overview of EDCA Analytical Model of EDCA Simulation Setup and Results Observations How to Handle Each Observation? Experimental Setup and Results Conclusion

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 30 Experiment setup Device and driver –WLAN devices with the Atheros chipset (e.g. Netgear WG511T) Basic chipset – most of the MAC functionality is handled in the driver –Linux-based MADWifi (Multiband Atheros driver for WiFi) driver –Implement the enhanced EDCA in the Hardware Access Layer (HAL) module HAL is similar to firmware and provides an interface to set some parameters, such as CW Experiment scenarios –AP estimates number of stations in each class, computes the optimal CW for each class, and broadcast these information in beacon messages. –One AP and two classes, one station in each class –Pre-specify bandwidth allocation ratio AP

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 31 Experimental results Figure 11. Throughputs attained by two traffic classes with on-off traffic. The pre-set ratio is r 1/2 = 4. Left: Throughputs (Mbps) Right: Throughput ratio.

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 32 Outline Challenges and Overview of IEEE s Mesh WLAN Overview of EDCA Analytical Model of EDCA Simulation Setup and Results Observations How to Handle Each Observation? Experimental Setup and Results Conclusion

doc.: IEEE /0364r0 Submission March 2006 Chunyu Hu, UIUCSlide 33 Conclusions Have analytically studied the impact of CW and AIFS in EDCA and how to apply it to IEEE s WLAN Insights obtained –Tuning AIFS (small value) has to be cautiously used, so as not to starve best-effort traffic –CW has to be tuned dynamically in response to varying load Can now apply these insights to provide per flow QoS or Prioritized QoS and effect congestion control in IEEE s MESH WLAN.