Dynamic Load Balancing through Association Control of Mobile Users in WiFi Networks 2013 YU-ANTL Seminal November 9, 2013 Hyun dong Hwang Advanced Networking Technology Lab. (YU-ANTL) Dept. of Information & Comm. Eng, Graduate School, Yeungnam University, KOREA (Tel : ; Fax :

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 2 Outline  INTRODUCTION  RELATED WORK AND IEEE BASICS Related Work IEEE Protocols  SYSTEM MODEL AND PROBLEM DEFINITION  DISTRIBUTED ASSOCIATION ALGORITHM  PERFORMANCE EVALUATION  CONCLUSION  Reference

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 3 INTRODUCTION  Wireless LAN shortcoming Comparing with wired line, the wireless channel is notorious for its instability owing to fading and losses. To be adaptive to the dynamic nature of wireless medium, data rate adaptation mechanisms such as Auto Rate Fallback (ARF) or Receiver Based Auto Rate (RBAR) are widely deployed for current WiFi products MAC has an “anomaly” that the throughput of high data rate MUs in good channel condition is down-equalized to that of the lowest data rate peer in the network. In this way, the data rate information is required to guide load balancing schemes.

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 4 INTRODUCTION  Example scenario Default best-RSSI(receiving signal strength indicator)-based AP selection scheme : Does not provide any fair sharing functionality will lead to unbalanced traffic load distribution. Bad MU-AP associations result in severe unfairness and even poor overall performance

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 5 INTRODUCTION  Centralized optimization Pros.  Collects information from the entire network, and then derive an optimal configuration based on complex computation. Con.  Such an approach is not scalable due to the NP-hard nature of the problem  Requires a separate processing infrastructure for performing the centralized computations.  Distributed heuristic methods Pros.  More flexible and they do not require the management center. Cons.  Some of the distributed AP selection schemes do not consider the multiple data rate information  Propose non-practical solutions

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 6 INTRODUCTION  New distributed heuristic algorithm Objective  Achieve load balancing by incorporating the multi-rate information Method  Add one additional field in the AP beacon and probing packets.  Some low complexity operations should be inserted in the Mus.

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 7 RELATED WORK AND IEEE BASICS  Centralized optimization techniques Most of the fine-grained for association control are based on centralized optimization techniques. Before applying the complex computation, researchers formulate the original problem for different fairness criteria  Eg.) Max-min fairness,Proportional fairness  Centralized schemes usually assume that all information about the network is already known beforehand to provide more accurate results.  However, it is not practical to let one control center collect all the information inside the network and distribute the association commands to the MUs.

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 8 RELATED WORK AND IEEE BASICS  Distributed schemes Propose a distributed selection scheme that balances the load according to the number of MUs associated with the APs. -> did not incorporate the multi-rate information. Recently propose distributed fair algorithms by incorporating the multi- rate information - > dependent on the specific features of not yet- deployed IEEE e.

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 9 RELATED WORK AND IEEE BASICS  IEEE Protocols Two access methods  Distributed Coordination Function(DCF) : control by station  Point Coordination Function(PCF) : control by AP DCF  Carrier Sense Multiple Access/Collision Avoidance (CSMA/CA) protocol  Provides two access schemes, the basic scheme and the request to send/clear to send (RTS/CTS) scheme.  Channel is sensed to be idle for a time interval equal to the DCF inter-frame space (DIFS), the MU simply transmits the packet.

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 10 SYSTEM MODEL AND PROBLEM DEFINITION  Condition Assume that the neighboring APs are configured with different non- overlapping channels Total 11 non-overlapping channels in IEEE a WLANs, the network manager can achieve this by careful frequency planning Assume that the traffic is saturated unidirectional UDP (user datagram protocol) packets with the same packet length throughout the network.  Denote Ap : A MU(Mobile User) : U Physical data rate of MU : r Packet length : L AP a load : Ya

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 11 SYSTEM MODEL AND PROBLEM DEFINITION  AP a is defined as the aggregate period of time AP a to provide a unit of traffic volume to all associated users u ∈ Ua.  MU throughput θu for all the MUs associated with AP a is  Time required to transmit one packet from MU u ∈ Ua  DATAu is the amount of time to transmit one data packet

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 12 DISTRIBUTED ASSOCIATION ALGORITHM  Association Algorithm for APs and Mus Legacy IEEE standard, the management packets from the AP do not contain any field indicating the AP load information. Add one additional field to the beacon and probing packets

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 13 DISTRIBUTED ASSOCIATION ALGORITHM  APs should keep updating the AP load by iterative moving average  If a MU is not associated with any AP in the network, it immediately scans all channels by sending probe request messages and receives response packets from the available Aps  Proposed AP selection strategy Least estimated load by supposing that it will be associated with all available AP

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 14 PERFORMANCE EVALUATION  Numerical evaluation based on the developed simulation program.  56 APs and 126 MUs with their mobility placed in a rectangle topology of size 1100×1000m2

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 15 PERFORMANCE EVALUATION  Numerical Simulation for Realistic Scenario Total throughput achieved by the proposed scheme

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 16 PERFORMANCE EVALUATION  Value of fairness metric

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 17 PERFORMANCE EVALUATION  Packet Level Simulation Simulation for a scenario (9 APs and 40 MUs) with 10 MUs suddenly roaming around AP1

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 18 PERFORMANCE EVALUATION  UDP or TCP traffic is applied separately for this scenario.  Configure UDP traffic with 2Mbps constant bit rate.

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 19 PERFORMANCE EVALUATION  Prototype Implementation

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 20 PERFORMANCE EVALUATION  The respective throughputs and total throughput were measured

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 21 CONCLUSION  Proposed a distributed and selfstabilized association scheme for the MUs in the multi-rate WLANs.  Proposed scheme gradually balances the AP loads in a distributed manner

Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) YU-ANTL Lab Seminal Hyun dong Hwang 22 Reference [1] Huazhi Gong, JongWon Kim, “Dynamic Load Balancing through Association Control of Mobile Users in WiFi Networks”, IEEE Transactions on Consumer Electronics, May 2008