Quality of Service Support in Wireless Networks Wireless Networks Laboratory (WINET) Quality of Service Support in Wireless Networks Hongqiang Zhai http://www.ecel.ufl.edu/~zhai Wireless Networks Laboratory Department of Electrical and Computer Engineering University of Florida In Collaboration with Dr. Xiang Chen and my advisor Professor Yuguang ``Michale’’ Fang

Wireless Networks Laboratory (WINET) Outline Introduction Performance analysis of the IEEE 802.11 MAC protocol A call admission and rate control scheme Conclusion and future research issues

Wireless Networks Laboratory (WINET) Wireless Landscape

Wireless Local Area Networks/ Wi-Fi Hot Spots Wireless Networks Laboratory (WINET) Wireless Local Area Networks/ Wi-Fi Hot Spots Web traffic Email Streaming video Instant messaging Gaming over IP Voice over IP over Wi-Fi Next Call May Come from a Wireless Hot Spot

Mobile Ad Hoc Networks and Wireless Mesh Networks Wireless Networks Laboratory (WINET) Mobile Ad Hoc Networks and Wireless Mesh Networks

Quality of Service (QoS) Requirements Wireless Networks Laboratory (WINET) Quality of Service (QoS) Requirements Bandwidth Delay and delay jitter Packet loss rate

Challenges Unreliable physical channel Wireless Networks Laboratory (WINET) Challenges Unreliable physical channel Time-varying propagation characteristics Interference Limited bandwidth Limited processing power and battery life Distributed control Mobility

Medium Access Control Coordinate channel access Reduce collision Wireless Networks Laboratory (WINET) Medium Access Control Coordinate channel access Reduce collision Efficiently utilize the limited wireless bandwidth B D C A

IEEE 802.11 Distributed Coordinate Function (DCF) MAC Protocol Wireless Networks Laboratory (WINET) IEEE 802.11 Distributed Coordinate Function (DCF) MAC Protocol Carrier sense multiple access with collision avoidance (CSMA/CA) Carrier sensing Physical Carrier Sensing Virtual Carrier Sensing Interframe Spacing (IFS) Short IFS (SIFS) < DCF IFS (DIFS) Binary Exponential Backoff Randomly chosen from [0, CW] CW doubles in case of collision Contention based MAC Can it support QoS requirements of various applications? Request to send DIFS DATA Transmitter Receiver Others B A ACK … Backoff NAV(RTS) RTS SIFS DATA SIFS SIFS NAV(CTS) CTS ACK Clear to send Acknowledge DIFS RTS … Backoff

Previous Work on Performance Analysis of the IEEE 802.11 MAC Standard Wireless Networks Laboratory (WINET) Previous Work on Performance Analysis of the IEEE 802.11 MAC Standard Previous studies focus on saturated case Each device always has packets in the system and keeps contending for the shared channel. Collision probability is very high Delay performance is very bad Only throughput and average delay have been derived. Related work Bianchi, JSAC March 2000 Cali et al., IEEE/ACM Tran. Networking, Dec. 2000 QoS requirements of real-time services can not be guaranteed if there are many contending users?

Previous Work on Supporting QoS in WLANs Wireless Networks Laboratory (WINET) Previous Work on Supporting QoS in WLANs Service differentiation Provide different channel access priorities for different services by differentiating Contention window Interframe spacing (IFS) IEEE 802.11e draft (based on 802.11b) Related work Ada and Castelluccia, Infocom’01 (CW, IFS) Veres et al., JSAC Oct. 2001 (real-time measurement in virtual MAC) S.T. Sheu and T.F. Sheu, JSAC Oct. 2001 (real-time traffic periods) S. Mangold et al., Wireless Communications Dec. 2003 (802.11e) Service differentiation is still not enough to meet the strict QoS requirements Can the IEEE 802.11 MAC protocol do better than service differentiation? Research issues Performance in both non-saturated and saturated case Probability distribution of medium access delay

MAC Service Time MAC service time is discrete in value Wireless Networks Laboratory (WINET) MAC Service Time Packet arrival MAC service time is discrete in value SIFS, DIFS, EIFS Backoff time is measured in time slots Packet to be transmitted is also discrete in length Transmit queue 1 2 3 2 3 3 MAC 2 3 1 Probability Generating Function (PGF) Pr{Ts=tsi}=pi (0 ≤ i < ∞)

MAC Service Time Widely used method Wireless Networks Laboratory (WINET) MAC Service Time start end Widely used method Calculate the average # of retransmissions NR = p/(1-p) Average transition time is NR × τ1 + τ2= Generalized state transition diagram (GSTD) Mark the PGF of the transition time on each branch along with the transition probability PGF of the transition time between two states is the corresponding system transfer function

MAC Service Time of IEEE 802.11 Wireless Networks Laboratory (WINET) MAC Service Time of IEEE 802.11 State variable (j, k): j is the backoff stage, k is the backoff timer Wj: the contention window at backoff stage j p: collision probability perceived by a node : maximum # of retransmissions

MAC Service Time of IEEE 802.11 Wireless Networks Laboratory (WINET) MAC Service Time of IEEE 802.11 Collision probability p payload size = 8000 bits, with RTS/CTS MAC service time (ms) MAC service time (ms) PDF Observation: When p is small, both the mean and standard deviation of MAC service time are small.

Delay and Delay Variation Wireless Networks Laboratory (WINET) Delay and Delay Variation Packet arrival TS Transmit queue 1 2 3 TW MAC TR

Network Throughput (Tidl pi ) (Tcol pc ) (Tsuc ps ) Wireless Networks Laboratory (WINET) Network Throughput (Tidl pi ) (Tcol pc ) (Tsuc ps ) Channel utilization: Normalized throughput : Channel busyness ratio: With RTS/CTS Without RTS/CTS n: # of nodes : the prob. that a node transmits in any slot

Wireless Networks Laboratory (WINET) Network Throughput Maximum throughput with good delay performance Collision Probability p Channel Busyness Ratio is an accurate, robust, and easily obtained sign of network status.

Wireless Networks Laboratory (WINET) Packet Loss Rate Given the collision probability p, the MAC layer may drop the packet with the probability Avg. queue length Pkt loss rate Channel busyness ratio

Model Validation The optimal operating point denoted by Umax Wireless Networks Laboratory (WINET) Model Validation Channel Busyness Ratio The optimal operating point denoted by Umax Simulation settings 50 nodes, RTS/CTS mechanism is used Each node has the same traffic rate. We monitor the performance at different traffic rates.

Call Admission and Rate Control (CARC) Wireless Networks Laboratory (WINET) Call Admission and Rate Control (CARC)

Call Admission Control Wireless Networks Laboratory (WINET) Call Admission Control Channel utilization/channel busyness ratio for a flow Admission control test R: flow data rate (bps) L: average packet length (bits) Up to Urt (= γ Umax, 0<γ<1) can be assigned to real-time traffic

Rate control Notation: r: Channel resource r allocated to each node Wireless Networks Laboratory (WINET) Rate control Notation: r: Channel resource r allocated to each node Allowable channel time occupation ratio tp: channel time for packet p Time that a successful transmission of packet p will last over the channel. ∆: scheduled interval Time between two consecutive packets that DRA passes to the MAC layer br: channel busyness ratio brth = Umax

Rate control Initialization Procedure: r=rstart Wireless Networks Laboratory (WINET) Rate control Initialization Procedure: r=rstart Three-Phase Resource Allocation Mechanism: multiplicative-increase if underloaded, i.e., br < BM=α×brth Additive-increase if moderately loaded, i.e., BM ≤ br < brth Multiplicative-decrease if heavily loaded, i.e., br ≥ brth

Theoretical Results of CARC Wireless Networks Laboratory (WINET) Theoretical Results of CARC Convergence of Multiplicative-Increase Phase

Theoretical Results of CARC Wireless Networks Laboratory (WINET) Theoretical Results of CARC Convergence to Fairness Equilibrium

Simulation Studies Simulation settings in ns2 Channel rate = 11 Mbps Wireless Networks Laboratory (WINET) Simulation Studies Simulation settings in ns2 Channel rate = 11 Mbps Voice traffic with an on-off model The on and off periods are exponentially distributed with an average value of 300 ms each. During on periods, traffic rate is 32kb/s with a packet size of 160 bytes. Greedy best effort traffic Saturated CBR traffic with a packet size of 1000bytes.

Throughput and MAC delay Wireless Networks Laboratory (WINET) Throughput and MAC delay Each node is a source of greedy traffic CARC improves the throughput by up to 71.62% with RTS/CTS, and by up to 157.32% without RTS/CTS CARC achieves up to 95.5% of maximum throughput with and without RTS/CTS

Fairness A new greedy node joins the network every other 10 seconds Wireless Networks Laboratory (WINET) Fairness Throughput Time (s) A new greedy node joins the network every other 10 seconds Higher aggregate throughput Throughput Time (s) Fairness convergence speed: 0-2 s Short term fairness

Quality of Service for Voice Traffic Wireless Networks Laboratory (WINET) Quality of Service for Voice Traffic 50 greedy nodes A new voice node joins the network every other 10 seconds. 97%ile 99%ile 0.0406 s 0.0811 s

Wireless Networks Laboratory (WINET) Conclusion The IEEE 802.11 MAC protocol can support strict QoS requirements of real-time services while achieving maximum throughput. Channel busyness ratio is a good network status indicator of the IEEE 802.11 systems. An efficient call admission and rate control framework is proposed to provide QoS for real-time service and also to approach the maximum throughput.