Reversing the Collision Avoidance Handshake in Wireless Networks J.J. Garcia-Luna-Aceves and Makis Tzamaloukas Computer and Communications Research Group (CCRG) Computer Engineering Department Jack Baskin School of Engineering University of California Santa Cruz, CA 95064

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas2 Presentation Outline n Sender-Initiated Collision Avoidance Protocols n Motivation n Polling Issues n Correct Collision Avoidance n RIMA protocols n Throughput and Delay Analysis n Conclusions

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas3 Sender-Initiated Protocols n SRMA (Kleinrock and Tobagi, August ‘76) n MACA (Karn, April ‘90) n MACAW (Bharghavan, Demers, Shenker and Zhang, August ‘94) n FAMA (Garcia-Luna-Aceves and Fullmer, September ‘97) n IEEE (July ‘97)

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas4 Motivation n The recipient of data packet is the point of interest n Recast the collision avoidance dialogues so that the receiver, sender or both can have control of the dialogue n Provide equal or better throughput than any sender-initiated IEEE like MAC protocol n Be applicable to multi-channel frequency-hopping or direct- sequence spread-spectrum radios n To date no receiver-initiated MAC protocol ensures correct collision avoidance n The receivers poll the senders

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas5 Ensuring Correct Collision Avoidance n First receiver-initiated MAC protocol was MACA-BI (Talucci and Gerla): – a node sends a DATA packet if it has previously received an RTR – a polled node can send a DATA packet either to the polling node or to any other neighbor – does not guarantee correct floor acquisition

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas6 Ensuring Correct Collision Avoidance n At time t0, nodes a and d send RTRs to b and e; at time t1 both b and e send DATA to c resulting in a collision abc d e t0 RTR to bRTR to e DATA to c t1 DATA to c

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas7 Ensuring Correct Collision Avoidance n At time t0, node a sends an RTR to b; at time t1, b sends DATA to a; at time t2 < t1 + , c sends an RTR to d; at time t3, d sends DATA to c; if data packets last longer than  + 3  the DATA from b and d collide at c abcd t0 RTR DATA t1 DATA t2 t3

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas8 Ensuring Correct Collision Avoidance n With RIMA-SP correct floor acquisition is guaranteed: – the polled node transmits a DATA packet only to the polling node – a collision avoidance period of  seconds is required at a polled node prior to answering an RTR; we have shown that  =  – an additional control signal is introduced; we call this signal No-Transmission-Request (NTR)

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas9 RIMA-SP illustrated RTR ZX XZ data RTR ZX XZ BACKOFF RTR ZX NTR ZX RTR data Successful handshake Colliding RTRs Hidden node interference

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas10 RIMA-SP timing diagrams RTR Waiting period DATA X Z RTRNTR BACKOFF X Z interference RTR collision X Z channel BACKOFF Node X sends an RTR and after  seconds receives a DATA packet Nodes X and Z send RTRs within  seconds and therefore a collision occurs Due to interference from node Z, node X sends an NTR to stop the handshake Noise detected at Z

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas11 Polling Issues n When to poll: whether or not the polling rate is independent of the data rate at polling nodes – independent polling – data driven polling n To whom: whether the poll is sent to a particular neighbor or to all neighbors; for dense networks a schedule must be provided to the poll recipients n How: whether the polling packet asks for permission to transmit as well

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas12 RIMA Protocols n Polling done with RTR (Request-To-Receive) packet n Three Receiver Initiated Medium Access (RIMA) protocols defined based on the type of polling: – RIMA-SP: A simple poll receiver initiated protocol (only the receiver sends data in a successful busy period) – RIMA-DP: A dual poll receiver initiated protocol (2 data packets are sent in the same successful busy period) – RIMA-BP: A broadcast poll receiver initiated protocol (the RTR is sent to everybody)

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas13 RIMA-DP illustrated RTR ZX XZ CTS data ZX RTR ZX XZ data ZX RTR ZX XZ BACKOFF RTR ZX NTR ZX data Success 1 Success 2 Failure 1 Failure 2

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas14 RIMA-DP timing diagrams RTRNTR Noise detected at Z BACKOFF X Z interference RTR Waiting period DATA CTS X Z RTR Waiting period DATA X Z RTR collision X Z channel BACKOFF Node X sends an RTR and after  seconds receives a DATA packet and then sends its DATA Node X sends an RTR and node Z replies with a CTS; node X sends its DATA Nodes X and Z send RTRs within  seconds and therefore a collision occurs Due to interference from node Z, node X sends an NTR to stop the handshake DATA

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas15 RIMA-BP illustrated RTR X Z ZX BACKOFF RTR ZX XZ RTSdata ZX RTR ZX Y XZ RTS NTR Z Y X RTS Success 1 Failure 1 Failure 2

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas16 RIMA-BP timing diagrams RTR collision X Z channel BACKOFF RTR Waiting period DATARTS X Z RTR RTS X Z Y collision NTR Node X sends an RTR; node Z responds with a RTS and after  seconds sends its DATA Node X sends an RTR; nodes Z and Y send an RTS within  seconds; node X sends an NTR to stop the handshake Nodes X and Z send RTRs within  seconds and therefore a collision occurs

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas17 Throughput Analysis Model n fully-connected network of N nodes n single, unslotted channel, error-free n the size for an RTR, RTS and CTS is  seconds; the size for a data packet is  seconds n the turn-around time is considered to be part of the duration of control and data packet n the propagation delay of the channel is  seconds n a polled node receiving an RTR always has a data packet to send n the probability that the packet is addressed to the polling node is 1/N

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas18 Throughput Analysis n 500 Byte data packets; 1Mbps network speed; maximum distance between nodes is 1 mile; on the left a 10 node network; on the right a 50 node network

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas19 Delay Analysis 500 Byte data packets; 1Mbps network speed; maximum distance between nodes is 1 mile

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas20 Heavy Traffic Approximation n MACA-BI analysis assumes that there is always a DATA packet after a successful poll n To present a fair comparison between MACA-BI and RIMA protocols we analyze a heavy load approximation where there is always a DATA packet to be sent after receiving an RTR

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas21 Throughput Analysis n 500 Byte data packets; 1Mbps network speed; maximum distance between nodes is 1 mile; network of 50 nodes; heavy load approximation throughput: any polled node always has data to send to any polling node

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas22 Conclusions n RIMA protocols provide correct collision avoidance in the presence of hidden terminals n The throughput achieved with RIMA-DP is higher than any other sender initiated MAC protocol for fully connected networks n RIMA-DP achieves higher throughput than all other collision avoidance protocols in fully-connected networks under heavy load approximation n Relative differences in performance remain the same in networks with hidden terminals

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas23 Simulation Results Base N1 N2 B1 B2 N1 (a) (b) (c) Base

MobiCom '99, August JJ. Garcia-Luna-Aceves and A. Tzamaloukas24 Simulation Results