Report

2014 YU-ANTL Lab Seminar Impact of Block ACK Window sliding on IEEE 802.11n throughput performance June 7, 2014 Shinnazar Seytnazarov Advanced Networking Technology Lab. (YU-ANTL) Dept. of Information & Comm. Eng, Graduate School, Yeungnam University, KOREA (Tel : +82-53-810-3940; Fax : +82-53-810-4742 http://antl.yu.ac.kr/; E-mail : seytnazarovsho@ynu.ac.kr) OUTLINE Introduction Frame aggregations BAW sliding The analytical model Expected A-MPDU length derivation Throughput derivation Analytical results Conclusion References Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 2 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Introduction (1) A-MPDU (Aggregation of MPDUs) - aggregation scheme [1] Sender can aggregate up to 64 MPDUs in A-MPDU frame If receiver receives at least one of the MPDUs successfully, it sends back Block ACK (Block acknowledgement) frame informing about transmission status MPDUs MPDU MAC delimiter header MAC payload FCS Pad MPDU PLCP header MPDU1 MPDU2 ... MPDU64 Tail/Pad A-MPDU Fig. 1. Aggregation of MPDUs Starting Sequence Number 1 2 3 ... 63 64 Bitmap PLCP header Frame Control Receiver Transmitter BA BA Address Address Control Info FCS Fig. 2. Block ACK frame format Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 3 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Introduction (2) Block ACK Window (BAW) sliding [1] BAW size is equal to 64 that is the maximum allowed A-MPDU length Sender can transmit the MPDUs that are within the BAW BAW continues sliding forward unless any of the MPDUs inside the BAW fails BAW sliding direction Previous position of BAW Current position of BAW 100 101 102 101 102 103 ... Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 4 163 164 165 166 167 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Introduction (3) Simple example for BAW = 4 Sender’s window Sender is sending SNs: 101~104 ... Transmitted and successfully received MPDU ... Transmitted but failed MPDU ... New MPDU in A-MPDU and BAW ... MPDU outside of BAW Receiver’s window Receiver is anticipating SNs: 101~104 101 102 103 104 105 101 102 103 104 105 ... Erro r 100 100 ... TX 104 103 102 101 A-MPDU1 Receiver is anticipating SNs: 103, 105, 106 TX 1 Sender is sending SNs: 103, 105, 106 103 104 105 106 107 0 1 102 103 104 105 106 107 ... Erro r 102 1 BlockACK1 ... TX 106 105 103 A-MPDU2 Receiver is anticipating SN: 103 TX 0 Sender is sending SN: 103 102 103 104 105 106 107 1 1 0 102 BlockACK2 103 104 105 106 107 ... ... TX 103 Receiver is anticipating SNs: 107~110 TX ACK Sender is sending SNs: 107~110 106 107 108 109 110 111 106 107 108 109 110 111 ... ... TX 110 Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 109 108 A-MPDU3 5 107 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Expected A-MPDU length derivation (1) We introduce several random variables: L – number of MPDUs in A-MPDU i.e. length of A-MPDU, L = 1, 2, . . , 64 N – number of new MPDUs in A-MPDU, N = 0, 1, 2, . . , L S – number of successful MPDUs in A-MPDU, S = 0, 1, 2, . . , L F – number of failed/erroneous MPDUs in A-MPDU, F = 0, 1, 2, . . , L X – number of successful MPDUs until the first failure in A-MPDU, X = 1, 2, . . , L We need to find: Expected number of MPDUs in A-MPDU - E[L] Expected number of successful MPDUs in A-MPDU - E[S] Expected number of failed MPDUs in A-MPDU - E[F] Assumptions: Sender’s buffer always has enough number of MPDUs to fill the BAW window MPDU errors occur independently and identically over MPDUs of A-MPDU Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 6 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Expected A-MPDU length derivation (2) [2] Considering assumption (2), the number of failed MPDUs has binomial distribution F ~ B(pe, L), where pe is MPDU error probability and L is the number of MPDUs in A-MPDU: P F=k = L k pke (1 − pe )L−k (1) So, the expected number of failed/erroneous MPDUs is: L k=0 k P EF = F=k = L L k=0 k k pke 1 − pe L−k (2) = pe L Number of successfully transmitted MPDUs also has a binomial distribution S ~ B(1 - pe, L): P S=n = L n (1 − pe )n pL−n e (3) So, the expected number of successful MPDUs per A-MPDU is: L n=0 n P ES = S=n = L L n=0 n n (1 − pe )n pL−n = 1 − pe L = L − E[F] e (4) PMF for the number of first successful MPDUs in A-MPDU can be written as: P X=k = (1 − pe )k pe , for 1 ≤ k ≤ L − 1 (1 − pe )k , for k = L ( 5) Using the above PMF we can calculate expected number of new MPDUs in A-MPDU; the expected window shift, where W depicts the window size which is 64: E[N] = L−1 W k=1 k L (1 − k k+1 W L−1 k=1 k(q −q ) L pe )k pe + L W L (1 − pe )L = L−1 W k k=1 k L q (1 − W L + WqL = q−qL+1 1−q gives q) + WqL = + WqL = q − q2 + 2q2 − 2q3 + ⋯ + L − 1 qL−1 − L − 1 qL q + q2 + q3 + ⋯ + qL−1 + qL − LqL − LqL W L + WqL W L + WqL = (6) Here, q = 1 − pe Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 7 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Expected A-MPDU length derivation (3) [2] The length of A-MPDU – L is the composition of failed MPDUs of previous A-MPDU and newly included MPDUs. L(i) = F(i − 1) + N(i) (7) It is obvious that under certain channel conditions, the expected length of AMPDU is the sum of the expectations of failed MPDUs and new MPDUs: E L =E F +E N (8) Thus, we will use the expected A-MPDU length instead of A-MPDU length for Equations (1-6): E L = E F + E N = pe E L + Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) q−qE L +1 1−q 8 − E L qE L W EL + WqE L = 1−qE L 1−q − YU-ANTL Lab. Seminar Shinnazar Seytnazarov Performance of BAW sliding under different channel conditions (1) Expected length of A-MPDU for different window sizes under different channel conditions E[L](W=64) E[L](W=128) 140 120 100 80 60 35.34 40 20.65 20 24.30 14.57 0 0 0.05 0.1 0.15 0.2 0.25 0.3 MPDU error probability Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 9 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Performance of BAW sliding under different channel conditions (2) Expected length of A-MPDU, expected number of successful and failed MPDUs under different channel conditions E[L](W=64) E[S](W=64) E[F](W=64) 70 60 50 40 30 20 10 0 0 0.05 0.1 0.15 0.2 0.25 0.3 MPDU error probability Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 10 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Discrete time Markov chain [3] Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 11 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Transmission probability Transmission probability τ that a station transmits in a randomly chosen slot time. τ= 2(1−2p) 1−2p w+1 +pw 1− 2p m (10) p is backoff stage increment probability due to either collision or A-MPDU failure because of channel noise: (11) p = 1 − (1 − τ)−1 1 − pe E L Equations (10) and (11) can be solved using numerical method and have unique solution for . Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 12 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Slot durations Idle slot duration Ti: When all STAs are counting down, no station transmits a frame and we have (12) Ti = σ Successful slot duration Ts: At least one MPDU in A-MPDU successfully received by receiver, the slot duration is the sum of a A-MPDU, a SIFS and an Block ACK duration Ts = TPHY_hdr + TA−MPDU + TSIFS + TBlock_ACK (13) Collision and ‘A-MPDU failure due to noise’ slot durations Tc and Tf: Tc = Tf = TPHY_hdr + TA−MPDU + TEIFS Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 13 (14) YU-ANTL Lab. Seminar Shinnazar Seytnazarov Probabilities of Time Slots Idle slot is observed if none of the stations transmits: Pi = (1 − τ)n (15) Successful slot is observed if only one station transmits and A-MPDU is not fully failed Ps = n 1 τ 1−τ n−1 1 − pe E L = nτ(1 − τ)n−1 1 − pe E L (16) Failure slot is observed if only one station transmits and A-MPDU is fully failed Pf = n 1 τ 1−τ n−1 p E L e = nτ(1 − τ)n−1 pe E L (17) Collision slot is observed if none of other slots is observed: (18) Pc = 1 − Pi − Ps − Pf Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 14 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Network throughput Network throughput can be defined as: S= E[successful A−MPDU length] E[slot duration] Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) = Ps E[S] Pk Tk 15 P 1−pe E[L] s Ts +Pc Tc +Pf Tf i i = P T +Ps (19) YU-ANTL Lab. Seminar Shinnazar Seytnazarov Parameters for numerical analysis MAC header MPDU payload PHY header duration Data transmission rate Block ACK transmission rate Maximum backoff stage (m) CWmin Slot duration (σ) DIFS SIFS Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 16 34B 1000B 44us 300/600Mbps 24Mbps 5 15 9us 28us 10us YU-ANTL Lab. Seminar Shinnazar Seytnazarov Performance analysis of IEEE 802.11n considering BAW sliding (1) Network throughput “with BAW” at R = 300Mbps Pe = 0.0 Pe = 0.1 Pe = 0.3 275 225 175 125 75 1 10 20 30 Number of stations Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 17 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Performance analysis of IEEE 802.11n considering BAW sliding (2) Network throughput “with BAW” at R = 600Mbps Pe = 0.0 Pe = 0.1 Pe = 0.3 500 400 300 200 100 1 10 20 30 Number of stations Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 18 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Performance analysis of IEEE 802.11n considering BAW sliding (3) Network throughput comparison “with and without BAW” at R = 300Mbps w/oBAW_Pe = 0.1 withBAW_Pe = 0.1 w/oBAW_Pe = 0.3 withBAW_Pe = 0.3 250 200 150 100 50 1 10 20 30 Number of stations Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 19 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Performance analysis of IEEE 802.11n considering BAW sliding (4) Network throughput comparison “with and without BAW” at R = 600Mbps w/oBAW_Pe = 0.1 withBAW_Pe = 0.1 w/oBAW_Pe = 0.3 withBAW_Pe = 0.3 450 400 350 300 250 200 150 1 10 20 30 Number of stations Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 20 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Performance analysis of IEEE 802.11n considering BAW sliding (5) Difference (%) between 'with BAW' and 'without BAW' at different PHY rates 600M_Pe = 0.3 300M_Pe = 0.3 600M_Pe = 0.1 300M_Pe = 0.1 60 50 40 30 20 10 0 1 10 20 30 Number of stations Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 21 YU-ANTL Lab. Seminar Shinnazar Seytnazarov Conclusion In this presentation We analyzed the BAW sliding effect on A-MPDU length under different channel conditions When MPDU error probability increases from 0.0 to 0.3 BAW decreases the AMPDU length from – – 64 to 14.57 for window size of 64 128 to 20.65 for window size of 128 BAW model was applied in DTMC model for IEEE 802.11n Network throughput was analyzed for different number of nodes and different channel conditions Existing DTMC models for IEEE 802.11n performance have huge difference: – – Over 20% when MPDU error probability 0.1 at 600Mbps PHY rate Over 10% when MPDU error probability 0.1 at 300Mbps PHY rate Conclusion BAW sliding has significant impact on A-MPDU size and network performance under erroneous channel conditions It is essential to consider BAW effect in order to have an accurate network performance estimations Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 22 YU-ANTL Lab. Seminar Shinnazar Seytnazarov References [1] IEEE 802.11n, Part 11: Standard for Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications Amendment 5: Enhancements for Higher Throughput, Sept. 2009. [2] Ginzburg, Boris, and Alex Kesselman. "Performance analysis of A-MPDU and A-MSDU aggregation in IEEE 802.11 n." In Sarnoff symposium, 2007 IEEE, pp. 1-5. IEEE, 2007. [3] G. Bianchi, “Performance analysis of the IEEE 802.11 distributed coordination function,” IEEE JSAC, vol. 18, no. 3, pp. 535–547, Mar. 2000. [4] T. Li, Q. Ni, D. Malone, D. Leith, Y. Xiao, and R. Turletti, “Aggregation with fragment retransmission for very high-speed WLANs,” IEEE/ACM Transactions on Networking, vol. 17, no. 2, pp. 591–604, Apr. 2009. [5] Chatzimisios, P., A. C. Boucouvalas, and V. Vitsas. "Influence of channel BER on IEEE 802.11 DCF." Electronics letters 39.23 (2003): 1687-9. Advanced Networking Tech. Lab. Yeungnam University (YU-ANTL) 23 YU-ANTL Lab. Seminar Shinnazar Seytnazarov