Accuracy Enhancements of the Model and EDCA QoS Extensions in ns-3

Transcription

1 Accuracy Enhancements of the Model and EDCA QoS Extensions in ns-3 Completion Talk Timo Bingmann Decentralized Systems and Network Services Research Group Institute of Telematics, June 26, 29

2 Roadmap 1 Thesis Objectives 2 Enhancements Propagation Loss Models Reception Criteria Frame Capture Effect EDCA Implementation 3 Speed Comparison 4 Conclusion Timo Bingmann - 2/19

3 Objectives 1 Thesis Objectives Compare implementations of new ns-3 network simulator with ns-2. Transfer extended ns-2 features added by the DSN to new ns-3 design. Implement EDCA extensions in ns-3. Evaluate performance gain of switching to ns-3. Timo Bingmann - 3/19

4 Constraints 1 Thesis Objectives All features must be thoroughly tested, evaluated and documented. Integrate cleanly into ns-3 design, which uses state-of-the-art software engineering methods. Researchers must be able to use them without detailed lower-layer knowledge. Timo Bingmann - 4/19

5 2 Enhancements Feature Comparison: ns-3.3 vs. ns-2.33 PHY Layer: No probabilistic Nakagami propagation model. Lacks modeling of frame capture effect. + BER/PER reception criterion for 82.11a. Results unequal to ns-2 s SINR criterion. MAC Layer: Support for EDCA extensions missing. + Overall good software design. Timo Bingmann - 5/19

6 2 Enhancements 2.1 Propagation Loss Models Nakagami Propagation Loss Model in ns-3 Ported Nakagami propagation loss model to ns-3. Extensively verified against ns-2 and the analytic probability density function. ns-2 ns-3 ns-2 Nakagami (defaults) ThreeLogDistance + Nakagami (default m =.75) ThreeLogDistance Probability Distance (m) -1 rxpower (dbm) Probability Distance (m) -1 rxpower (dbm) Timo Bingmann - 6/19

7 2 Enhancements 2.2 Reception Criteria Reception Criteria: SINR Implemented ns-2 s SINR reception criterion in ns-3 as Ns2ExtWifiPhy. B distance A ns-2 ns-3 Reception Probability FreeSpace TwoRayGround Nakagami (Log Only) Nakagami Defaults Nakagami-1 Nakagami-3 Nakagami-5 Reception probability Friis LogDistance (defaults) LogDistance (exponent = 2.2) ThreeLogDistance (defaults) ThreeLogDistance + Nakagami (defaults) ThreeLogDistance + Nakagami (m = 1.) ThreeLogDistance + Nakagami (m = 3.) ThreeLogDistance + Nakagami (m = 5.) Distance (m) Distance (m) Timo Bingmann - 7/19

8 2 Enhancements 2.2 Reception Criteria Discussion of SINR and BER/PER Detailed explanation of existing BER/PER reception in ns-3. Discussion and comparison against SINR. Packet Error Rate (PER) Free-space Reception Range Probability of packet error Pper Mb/s 9 Mb/s 12 Mb/s 18 Mb/s 24 Mb/s 36 Mb/s 48 Mb/s 54 Mb/s Reception probability Ns2Ext at 6 or 9 Mb/s Yans at 6 Mb/s Yans at 9 Mb/s Ns2Ext at 12 or 18 Mb/s Yans at 12 Mb/s Yans at 18 Mb/s Ns2Ext at 24 or 36 Mb/s Yans at 24 Mb/s Yans at 36 Mb/s Ns2Ext at 48 or 54 Mb/s Yans at 48 Mb/s Yans at 54 Mb/s SINR per bit γb (db) Distance (m) Timo Bingmann - 8/19

9 Frame Capture Effect Added frame capture effect to Ns2ExtWifiPhy. Evaluated against ns-2. 5 ns-2 2 Enhancements 2.3 Frame Capture Effect 5 B A B fixed t ns-3 C varying A Time Packet delay t (µs) Packet delay t (µs) Impossible due to CSMA/CA Received Distance between nodes C and A (m) 1 5 Impossible due to CSMA/CA Received always Received with preamble capture Received with data capture Distance between nodes C and A (m) Timo Bingmann - 9/19

10 Frame Capture Effect Added frame capture effect to Ns2ExtWifiPhy. Evaluated against ns-2. 5 ns-2 2 Enhancements 2.3 Frame Capture Effect 5 B A B fixed t ns-3 C varying A Time Packet delay t (µs) Packet delay t (µs) Impossible due to CSMA/CA Received Distance between nodes C and A (m) Distance between nodes C and A (m) Timo Bingmann - 9/19

11 EDCA Implementation 2 Enhancements 2.4 EDCA Implementation Extended ns-3 with EDCA capabilities. Builds up on the well designed DCF classes. Added TXOP limits and burst sequences. Tested individual maximum throughput against analytical reference values. Experiment with differently prioritized traffic streams shows relative QoS. Timo Bingmann - 1/19

12 2 Enhancements 2.4 EDCA Implementation QosAdhocWifiMac WifiQosTag AC: int AC VO AC VI AC BE AC BK DcaTxop DcaTxop DcaTxop DcaTxop AIFSN: int Backoff: int AIFSN: int Backoff: int AIFSN: int Backoff: int AIFSN: int Backoff: int Queue Queue Queue Queue Dcfmanager NAV MacLow SIFS: Time SlotTime: Time CCA BUSY WifiPhy WifiChannel Timo Bingmann - 11/19

13 2 Enhancements 2.4 EDCA Implementation Maximum Throughput Experiment DATA AIFS CW Frame DATA Without ACK AIFS CW Time AIFS CW DATA SIFS Time ACK Frame With ACK DATA SIFS AIFS CW DATA SIFS DATA SIFS DATA Time Frame TXOPLimit Superframe TXOP burst without ACKs DATA SIFS ACK SIFS DATA SIFS ACK SIFS DATA SIFS ACK AIFS CW Time Frame TXOPLimit Superframe TXOP burst with ACKs Timo Bingmann - 12/19

14 2 Enhancements 2.4 EDCA Implementation Maximum Throughput Experiment Reference value in B/s and relative difference of experimental result with 99 % error margin for 54 Mb/s data rate. 8 B - noack 8 B - ACK 234 B - ACK DCF ±.11.1 ±.1.1 ±.4 AC VO p/D4.2.3 ±.5.1 ±.2.1 ±.1 AC BK p/D4.2.6 ±.1.2 ±.9.1 ±.4 Tested 216 configurations. Maximum relative difference was.85 ±.11. Timo Bingmann - 13/19

15 2 Enhancements 2.4 EDCA Implementation EDCA Traffic Streams Experiment Payload rate received at listener (Mb/s) Without ACK AC VO AC VI AC BE AC BK Number of sending nodes Payload rate received at listener (Mb/s) With ACK AC VO AC VI AC BE AC BK Number of sending nodes Each node sends four 16 Kb/s streams with different ACs. As the number of nodes increases the medium is saturated. Timo Bingmann - 14/19

16 3 Speed Comparison Speed Comparison Highway Scenario Modeled identically in both ns-2 and ns-3. Made possible with newly added components. Timo Bingmann - 15/19

17 3 Speed Comparison Speed Comparison Results Packets sent (in thousands) Number of nodes Packets received (in millions) Number of nodes ns-2 unoptimized ns-2 optimized ns-2 icc optimized ns-3 debug ns-3 optimized ns-3 optimized static ns-3 icc optimized ns-3 icc optimized static ns-3 32-bit optimized ns-3 32-bit optimized static ns-2 nakagami optimized ns-3 nakagami optimized static Timo Bingmann - 16/19

18 3 Speed Comparison Speed Comparison Results Simulation run time (seconds) ns-2 unoptimized ns-2 optimized ns-2 icc optimized ns-3 debug ns-3 optimized ns-3 optimized static ns-3 icc optimized ns-3 icc optimized static ns-3 32-bit optimized ns-3 32-bit optimized static ns-2 nakagami optimized ns-3 nakagami optimized static Number of nodes Timo Bingmann - 17/19

19 3 Speed Comparison Speed Comparison Results Slowest configuration: ns-3 in debug mode. ns-3 optimized mode gives 76.3±.5% reduction. ns-3 optimized with static linking yields further reduction of 42.6±1.2%. Compilation without -fpic yielded a reduction of only 1.1±.3%. icc vs. gcc: no improvement, even slight speed decrease (1.9±.4%). Speed increase of ns-3 over identical ns-2 simulation: 58.6±1.8%. Enabling Nakagami propagation increases run time by 8.1±1.% in ns-3 and 3.8±.4% in ns-2. Timo Bingmann - 18/19

20 Conclusion 4 Conclusion Extended ns PHY layer to show equivalent behavior as ns-2. Improved MAC layer with EDCA extensions. All enhancements thoroughly verified. Speed test of ns-3 shows up to 59 % execution time reduction over ns-2. Thank you for your attention. Timo Bingmann - 19/19

21 Appendix 5 Enlarged Plots and Figures 5 Enlarged Plots and Figures Propagation Loss Models Reception Criteria Frame Capture Effect EDCA Implementation Timo Bingmann - 2/19

22 5 Enlarged Plots and Figures 5.1 Propagation Loss Models Object WifiChannel m_next PropagationLossModel CalcRxPower(txPowerDbm: double, a: Ptr<MobilityModel>, b: Ptr<MobilityModel>): double SetNext(next: Ptr<PropagationLossModel>): void DoCalcRxPower(txPowerDbm: double, a: Ptr<MobilityModel>, b: Ptr<MobilityModel>): double FriisPropagationLossModel m_lambda: double m_systemloss: double DoCalcRxPower(txPowerDbm: double, a: Ptr<MobilityModel>, b: Ptr<MobilityModel>): double ThreeLogDistancePropagationLossModel m_distance: double m_distance1: double m_distance2: double m_exponent: double m_exponent1: double m_exponent2: double m_referenceloss: double DoCalcRxPower(txPowerDbm: double, a: Ptr<MobilityModel>, b: Ptr<MobilityModel>): double NakagamiPropagationLossModel m_distance1: double m_distance2: double m_m: double m_m1: double m_m2: double m_erlangrandomvariable: ErlangVariable m_gammarandomvariable: GammaVariable DoCalcRxPower(txPowerDbm: double, a: Ptr<MobilityModel>, b: Ptr<MobilityModel>): double RandomPropagationLossModel m_variable: RandomVariable DoCalcRxPower(txPowerDbm: double, a: Ptr<MobilityModel>, b: Ptr<MobilityModel>): double Timo Bingmann - 21/19

23 5 Enlarged Plots and Figures 5.1 Propagation Loss Models ns-2 Nakagami Reception Power ns-2 Nakagami (defaults) Probability rxpower (dbm) Distance (m) 25 Timo Bingmann - 22/19

24 5 Enlarged Plots and Figures 5.1 Propagation Loss Models ns-3 NakagamiPropagationLossModel Nakagami (default m =.75) Probability rxpower (dbm) Distance (m) Timo Bingmann - 23/19

25 5 Enlarged Plots and Figures 5.1 Propagation Loss Models ns-3 ThreeLogDistance and Nakagami ThreeLogDistance + Nakagami (default m =.75) ThreeLogDistance Probability rxpower (dbm) Distance (m) Timo Bingmann - 24/19

26 5 Enlarged Plots and Figures 5.2 Reception Criteria PER for Different Modes Probability of packet error Pper Mb/s 9 Mb/s 12 Mb/s 18 Mb/s 24 Mb/s 36 Mb/s 48 Mb/s 54 Mb/s SINR per bit γ b (db) Timo Bingmann - 25/19

27 5 Enlarged Plots and Figures 5.2 Reception Criteria Two Nodes Experiment Scenario B distance A Timo Bingmann - 26/19

28 5 Enlarged Plots and Figures 5.2 Reception Criteria ns-2 Two Nodes Reception Range Reception Probability FreeSpace TwoRayGround Nakagami (Log Only) Nakagami Defaults Nakagami-1 Nakagami-3 Nakagami Distance (m) Timo Bingmann - 27/19

29 5 Enlarged Plots and Figures 5.2 Reception Criteria ns-3 Two Nodes Reception Range Reception probability Friis LogDistance (defaults) LogDistance (exponent = 2.2) ThreeLogDistance (defaults) ThreeLogDistance + Nakagami (defaults) ThreeLogDistance + Nakagami (m = 1.) ThreeLogDistance + Nakagami (m = 3.) ThreeLogDistance + Nakagami (m = 5.) Distance (m) Timo Bingmann - 28/19

30 5 Enlarged Plots and Figures 5.2 Reception Criteria ns-3 Mixed PHY Models Free-Space 1 Reception probability Ns2Ext at 6 or 9 Mb/s Yans at 6 Mb/s Yans at 9 Mb/s Ns2Ext at 12 or 18 Mb/s Yans at 12 Mb/s Yans at 18 Mb/s Ns2Ext at 24 or 36 Mb/s Yans at 24 Mb/s Yans at 36 Mb/s Ns2Ext at 48 or 54 Mb/s Yans at 48 Mb/s Yans at 54 Mb/s Distance (m) Timo Bingmann - 29/19

31 5 Enlarged Plots and Figures 5.2 Reception Criteria ns-3 Mixed PHY Models Nakagami Reception probability Ns2Ext at 6 or 9 Mb/s Yans at 6 Mb/s Yans at 9 Mb/s Ns2Ext at 12 or 18 Mb/s Yans at 12 Mb/s Yans at 18 Mb/s Ns2Ext at 24 or 36 Mb/s Yans at 24 Mb/s Yans at 36 Mb/s Ns2Ext at 48 or 54 Mb/s Yans at 48 Mb/s Yans at 54 Mb/s Distance (m) Timo Bingmann - 3/19

32 5 Enlarged Plots and Figures 5.2 Reception Criteria ns-3 Mixed PHY Models Nakagami m =1 Reception probability Ns2Ext at 6 or 9 Mb/s Yans at 6 Mb/s Yans at 9 Mb/s Ns2Ext at 12 or 18 Mb/s Yans at 12 Mb/s Yans at 18 Mb/s Ns2Ext at 24 or 36 Mb/s Yans at 24 Mb/s Yans at 36 Mb/s Ns2Ext at 48 or 54 Mb/s Yans at 48 Mb/s Yans at 54 Mb/s Distance (m) Timo Bingmann - 31/19

33 5 Enlarged Plots and Figures 5.3 Frame Capture Effect Three Nodes Capture Experiment B fixed C varying A B A t Time Timo Bingmann - 32/19

34 5 Enlarged Plots and Figures 5.3 Frame Capture Effect ns-2 Three Nodes Capture Packet delay t (µs) Impossible due to CSMA/CA Received Distance between nodes C and A (m) Timo Bingmann - 33/19

35 5 Enlarged Plots and Figures 5.3 Frame Capture Effect ns-3 Three Nodes Capture Packet delay t (µs) Impossible due to CSMA/CA Received always Received with preamble capture Received with data capture Distance between nodes C and A (m) Timo Bingmann - 34/19

36 5 Enlarged Plots and Figures 5.3 Frame Capture Effect ns-3 Three Nodes Capture Nakagami Packet delay t (µs) Distance between nodes C and A (m) Timo Bingmann - 35/19

37 5 Enlarged Plots and Figures 5.4 EDCA Implementation Maximum Throughput Experiment Reference value and difference of experimental result in B/s with 99 % error margin for 54 Mb/s data rate. 8 B - noack 8 B - ACK 234 B - ACK DCF ± ± ± AC VO p/D ± ± ± 39 AC BK p/D ± ± ± Tested 216 configurations. Maximum difference was 71 B/s ± Timo Bingmann - 36/19

38 5 Enlarged Plots and Figures 5.4 EDCA Implementation EDCA Traffic Streams no ACK Payload rate received at listener (Mb/s) AC VO AC VI AC BE AC BK Number of sending nodes Timo Bingmann - 37/19

39 5 Enlarged Plots and Figures 5.4 EDCA Implementation EDCA Traffic Streams with ACK Payload rate received at listener (Mb/s) AC VO.2 AC VI AC BE AC BK Number of sending nodes Timo Bingmann - 38/19