Study on Window-Based Reliable Multicast Protocols for Wireless LANs Huei-Wen Ferng, Ph.D. Assistant Professor Department of Computer Science and Information Engineering (CSIE) Nation Taiwan University of Science and Technology (NTUST) Wireless Communications and Networking Engineering (WCANE) Lab

9/6/2004 PIMRC 2004NTUST/WCANE Lab2 Outline Introduction Description of the proposed protocols Performance study and numerical examples Conclusions

9/6/2004 PIMRC 2004NTUST/WCANE Lab3 Unicast vs. Multicast Wired vs. Wireless Unreliable vs. Reliable We deal with the issue of incorporating the reliability into the multicast of wireless LANs. Two major problems: ACK/NAK implosion and media access Introduction (1/2)

9/6/2004 PIMRC 2004NTUST/WCANE Lab4 Introduction (2/2) Existing approaches (Kuri and Kasera [6]):  Delay feedback-based protocol (DBP)  Probabilistic feedback-based protocol (PBP)  Leader-based protocol (LBP) Based on LBP, we further propose  LBP with a sliding window (LBPW)  LBP with a sliding window and n-fold acknowledgement reduction (LBPR(n)) To achieve reliability, automatic repeat request (ARQ) is applied in this paper.

9/6/2004 PIMRC 2004NTUST/WCANE Lab5 Outline Introduction Description of the proposed protocols Performance study and numerical examples Conclusions

9/6/2004 PIMRC 2004NTUST/WCANE Lab6 Scenario Basic network architecture: AP and several mobile hosts We split the communication link into  Sender to APs  An AP to group members (GMs) Merits of such an arrangement  Scalability  Local error recovery LBPW and LBPR(n) are designed for the basic network architecture.

9/6/2004 PIMRC 2004NTUST/WCANE Lab7  Event PCW1 - AP to GMs (starting in slot k):  Send an RTS to all GMs  Event PCW2 - Leader/GMs to AP (in slot k+1):  Leader Send a CTS if it is ready to receive data frames; otherwise, do nothing.  Other GMs Send an NCTS if it is not ready to receive data frames; otherwise, do nothing. LBPW Phase of RTS/CTS exchange

9/6/2004 PIMRC 2004NTUST/WCANE Lab8  Event PTW1 - AP to GMs (in slot k+2):  If a CTS was received by the AP in slot k + 1, start to transmit contiguously available n a (<= WS) data frames with labels, say, 1, 2,..., n a ; otherwise, go back to event PCW1.  Event PTW2 - Leader/GMs to AP ( during slot k ⌈ (f l * t tr * n a + t pc + t pp )/t st ⌉ and slot k ⌈ (f l * t tr * n a + t pc + t pp )/t st ⌉ + n a ) :  Leader If the leader received the i th frame correctly, it sends an ACK in slot k+1+ ⌈ (f l * t tr * n a + t pc + t pp )/t st ⌉ +i ; otherwise, it sends a NAK. LBPW Phase of data frames transfer

9/6/2004 PIMRC 2004NTUST/WCANE Lab9  Other GMs If the ith frame was received with error bits by any GM, it sends a NAK in slot k ⌈ (f l * t tr * n a + t pc + t pp )/t st ⌉ + i ; otherwise, it does nothing.

9/6/2004 PIMRC 2004NTUST/WCANE Lab10 LBPW Based on feedbacks from GMs, AP should make a decision. Three cases AP faces:  An ACK is received  Nothing is received  A collision occurs Case I: frame is correctly received. Other cases: retransmission is required.

9/6/2004 PIMRC 2004NTUST/WCANE Lab11  Event PCW1 - AP to GMs (starting in slot k):  Send an RTS to all GMs  Event PCW2 - Leader/GMs to AP (in slot k+1):  Leader Send a CTS if it is ready to receive data frames; otherwise, do nothing.  Other GMs Send an NCTS if it is not ready to receive data frames; otherwise, do nothing. LBPR(n) Phase of RTS/CTS exchange

9/6/2004 PIMRC 2004NTUST/WCANE Lab12 LBPR(n) Phase of data frames transfer  Event PTR1 - AP to GMs (in slot k+2):  If a CTS was received by the AP in slot k + 1, start to transmit contiguously available n a (<= WS) data frames with labels, say, 1, 2,..., n a ; otherwise, go back to event PCW1.  Event PTR2 - Leader/GMs to AP ( during slot k ⌈ (f l * t tr * n a + t pc + t pp )/t st ⌉ and slot k ⌈ (f l * t tr * n a + t pc + t pp )/t st ⌉ + ⌈ n a /n ⌉ ) :  Leader Send an acknowledgement in a bit map, including the receiving status for at most n frames at a time. Hence ⌈ n a /n ⌉ times of ACKs are

9/6/2004 PIMRC 2004NTUST/WCANE Lab13 required to send during slot k+2+ ⌈ (f l * t tr * n a + t pc + t pp )/t st ⌉ and slot k+1+ ⌈ (f l * t tr * n a + t pc + t pp )/t st ⌉ + ⌈ n a /n ⌉.  Other GMs Break the n a frames into ⌈ n a /n ⌉ subsegments (each including exactly n frames except the last one). If one of frames for subsegment i was received with error bits by any GM, it sends a NAK directly in slot k ⌈ (f l * t tr * n a + t pc + t pp )/t st ⌉ + i ; otherwise, it does nothing.

9/6/2004 PIMRC 2004NTUST/WCANE Lab14 LBPR(n) Based on feedbacks from GMs, AP should make a decision. Two cases AP faces:  An ACK in a bit map is received  A collision occurs Case I: erroneous frames are retransmitted. Case II: all frames are retransmitted.

9/6/2004 PIMRC 2004NTUST/WCANE Lab15 Outline GSM/GPRS system Description of the proposed protocols Performance study and numerical examples Conclusions

9/6/2004 PIMRC 2004NTUST/WCANE Lab16 Assumptions A minimal wireless LAN is considered. MAC is neglected. Perfect time synchronization is assumed. One multicast group is considered. Each GM is assumed to be always ready to receive data frames. Data frames may be corrupted but not lost. Control frames are always correctly received. Frames are generated according to Batch Poisson.

9/6/2004 PIMRC 2004NTUST/WCANE Lab17 Performance Metrics Cost the average time lasting since the AP contends the channel until the AP ascertains that all group members correctly receive the frame. Exposure ratio of the number of mobile hosts actually receiving the frame and the number of mobile hosts who do need the frame. Average queueing delay/queue length, Number of ACKs or NAKs.

9/6/2004 PIMRC 2004NTUST/WCANE Lab18 LBPW vs. LBP (1/2) The reductions when WS = 2 and WS = 10 compared to LBP are 4.3 % and 7.0 %, when fl = 20 and nGM = % and 13.3 %, when fl = 10 and nGM = 10 The increase of the window size WS causes a lower cost, i.e., higher throughput. These results evidently show that LBPW with a large window size, say 10, performs much better than LBP.

9/6/2004 PIMRC 2004NTUST/WCANE Lab19 LBPW vs. LBP (2/2)  Exposure is not affected by the increase of the window size but it increases as the group grows up.  We see that the queueing delay goes down as the window size increases or the number of group members decreases.

9/6/2004 PIMRC 2004NTUST/WCANE Lab20 LBPR( n ) vs. LBPW (LBP) (1/3) LBPR(n) achieves ACKs/NAKs reduction approximately by a factor of n compared to LBP or LBPW.

9/6/2004 PIMRC 2004NTUST/WCANE Lab21 LBPR( n ) vs. LBPW (LBP) (2/3)

9/6/2004 PIMRC 2004NTUST/WCANE Lab22 LBPR( n ) vs. LBPW (LBP) (3/3) LBPR(n) performs better than LBPW due to the saving of ACKs/NAKs. The cost reduction is more obviously when FEP is high.

9/6/2004 PIMRC 2004NTUST/WCANE Lab23 Outline GSM/GPRS system Description of the proposed protocols Performance study and numerical examples Conclusions

9/6/2004 PIMRC 2004NTUST/WCANE Lab24 Conclusions The cost of LBPR(n) is lower than that of LBPW which is subsequently lower than LBP. The attainable cost reduction of LBPW compared to LBP can be over 10%. Both LBPW and LBPR(n) perform better than LBP in terms of queueing delay. LBPW mostly performs better than LBPR(n) for n ≥3, while LBPR(2) performs better than LBPW when the frame loss probability is low. As for the exposure metric, LBPW is the same as LBP and smaller than LBPR(n). For larger n, the exposure of LBPR(n) becomes higher. So, we suggest LBPW and LBPR(2) to be used.

9/6/2004 PIMRC 2004NTUST/WCANE Lab25 Thank You!