MAC Protocols for Sensor Networks

1. MAC Protocols for Sensor Networks Mahesh Arumugam (arumugam AT cse.msu.edu) Research Group Meeting February 16 2004 S-MAC: Medium Access Control with Coordinated, Adaptive Sleeping for Wireless Sensor Networks. W. Ye, J. Heidemann, and D. Estrin. To Appear in the IEEE/ACM Transactions on Networking. T-MAC: An Adaptive Energy-Efficient MAC Protocol for Wireless Sensor Networks. T. van Dam, and K. Langendoen. SenSys 2003. Hi everyone. Today, I will present some techniques proposed for local self-healing or local self-stabilization in wireless sensor networks, esp., for routing and clustering. Let me first introduce the concept of local self-stabilization or healing. Hi everyone. Today, I will present some techniques proposed for local self-healing or local self-stabilization in wireless sensor networks, esp., for routing and clustering. Let me first introduce the concept of local self-stabilization or healing.

2. Feb 16 2004 MAC for Sensor Networks 2 Guidelines for Good MAC Protocol Design Energy-efficiency protocol overhead, idle-listening, overhearing, collisions Scalability Adaptability Secondary concerns (mostly, appl. dependent): Fairness Latency Bandwidth utilization, …

3. Feb 16 2004 MAC for Sensor Networks 3 MAC Options for Sensor Networks Guaranteed message delivery TDMA, FDMA, CDMA Prime, LEACH, SS-S-TDMA Probabilistic guarantees on message delivery {S|T|B}-MAC Whenever faults such as node/link failures, state corruptions occur in wireless networks, the state of large number of nodes are affected and this can typically include all the nodes in the network. Such unbounded fault-propagation decreases the availability, stability and scalability of the system. Ideally, we would like faults to be contained around the regions where they occur. Also, we would like the time taken by the system to stabilize starting the state where faults occurred to be proportional to fault-affected/perturbed regions. We call this F-local stabilization. Consider the routing protocols in wireless networks. They can be classified as either link-state or distance-vector based protocols. In the link-state based protocols, all nodes maintain topological information. Hence, if a change occurs, all nodes need to know that. Therefore, F-local stabilization is not possible. In distance-vector based protocols such as routing-information protocol or the BGP protocol in the internet, nodes maintain only local information. Hence, F-local stabilization is possible.Whenever faults such as node/link failures, state corruptions occur in wireless networks, the state of large number of nodes are affected and this can typically include all the nodes in the network. Such unbounded fault-propagation decreases the availability, stability and scalability of the system. Ideally, we would like faults to be contained around the regions where they occur. Also, we would like the time taken by the system to stabilize starting the state where faults occurred to be proportional to fault-affected/perturbed regions. We call this F-local stabilization. Consider the routing protocols in wireless networks. They can be classified as either link-state or distance-vector based protocols. In the link-state based protocols, all nodes maintain topological information. Hence, if a change occurs, all nodes need to know that. Therefore, F-local stabilization is not possible. In distance-vector based protocols such as routing-information protocol or the BGP protocol in the internet, nodes maintain only local information. Hence, F-local stabilization is possible.

4. Feb 16 2004 MAC for Sensor Networks 4

5. Feb 16 2004 MAC for Sensor Networks 5 S-MAC: Overview Periodic listen and sleep Collision avoidance Coordinated sleeping Overhearing avoidance

6. Feb 16 2004 MAC for Sensor Networks 6 S-MAC: Periodic Listen & Sleep Frame Duty cycle (Listen Interval / Frame Length) Frame schedule Nodes are free to choose their listen/sleep schedule Requirement: neighboring nodes synchronize together Exchange schedules periodically (SYNC packet) Synchronization period (SP) Nodes communicate in receivers scheduled listen times

7. Feb 16 2004 MAC for Sensor Networks 7 S-MAC: Collision-Avoidance Collision-Avoidance Strategy ~= 802.11 RTS/CTS Physical carrier sense Virtual carrier sense: network allocation vector (NAV)

8. Feb 16 2004 MAC for Sensor Networks 8 S-MAC: Coordinated Sleeping (1) Frame Schedule Maintenance Choosing a schedule Listen to the medium for at least SP Nothing heard, choose a schedule Broadcast a SYNC packet (should contend for medium) Following a schedule Receives a schedule before choosing/announcing Follows the schedule Broadcast a SYNC packet Adopting multiple schedules Receives a schedule after choosing/announcing Can discard the new schedule; or Follow both the schedules – suffer more energy loss

9. Feb 16 2004 MAC for Sensor Networks 9 S-MAC: Coordinated Sleeping (2) Neighbor Discovery chance of failing to discover an existing neighbor corrupted SYNC packet, collisions, interference sensor – border of two schedules; discovers only the first schedule, if schedules do not overlap Periodically, listen for the complete SP frequency? ? - if a sensor has no neighbors S-MAC experimental values: SP = 10 seconds Neighbor discovery period = 2 minutes, if at least 1 nbr

10. Feb 16 2004 MAC for Sensor Networks 10 S-MAC: Coordinated Sleeping (3) Maintaining Synchronization Clock drifts – not a major concern (listen time = 0.5s – 105 times longer than typical drift rates) Need to mitigate long term drifts – schedule updating using SYNC packet (sender ID, its next scheduled sleep time – relative); Listen is split into 2 parts – for SYNC and RTS/CTS Once RTS/CTS is established, data sent in sleep interval

11. Feb 16 2004 MAC for Sensor Networks 11 S-MAC: Coordinated Sleeping (4) Example Scenarios

12. Feb 16 2004 MAC for Sensor Networks 12 S-MAC: Coordinated Sleeping (5) Adaptive Listening – Low-duty cycle to active mode * Overhearing nodes – wakeup at the end of the current transmission (duration field in RTS/CTS)

13. Feb 16 2004 MAC for Sensor Networks 13 S-MAC: Overhearing Avoidance (1) Who should sleep when a node is transmitting?

14. Feb 16 2004 MAC for Sensor Networks 14 S-MAC: Efficient Message Passing (1) Sending a long message? As a single packet: ? cost of re-transmission for message corruption

15. Feb 16 2004 MAC for Sensor Networks 15 S-MAC: Efficient Message Passing (2) RTS/CTS/ACK – has duration fields in it If ACK is not received, increase the transmission time, retransmit. ACK will be also be updated. Difference between 802.11 & S-MAC Medium is reserved upfront for the whole transmission in S-MAC

16. Feb 16 2004 MAC for Sensor Networks 16

17. Feb 16 2004 MAC for Sensor Networks 17 Drawbacks of S-MAC Active (Listen) interval – long enough to handle to highest expected load If message rate is less – energy is still wasted in idle-listening S-MAC fixed duty cycle – is NOT OPTIMAL

18. Feb 16 2004 MAC for Sensor Networks 18 T-MAC: Preliminaries Adaptive duty cycle: A node is in active mode until no activation event occurs for time TA Periodic frame timer event, receive, carrier sense, send-done, knowledge of other transmissions being ended Communication ~= S-MAC/802.11 Frame schedule maintenance ~= S-MAC

19. Feb 16 2004 MAC for Sensor Networks 19 T-MAC: RTS Operation (1) Contention Interval waiting/listening for a random time within a fixed contention interval (unlike exponential back-off in 802.11) Tuned for max. load assumptions: load is always high, does not vary

20. Feb 16 2004 MAC for Sensor Networks 20 T-MAC: RTS Operation (2) RTS Retries No CTS reply for RTS? collision receiver should not reply due to another transmission in progress (overhearing RTS/CTS of others) receiver is sleeping Solutions: wait for TA, go to sleep – receiver might be awake,and start transmission! retransmit RTS if no answer, max of 2 retries

21. Feb 16 2004 MAC for Sensor Networks 21 T-MAC: Choosing TA Requirement: a node should not sleep while its neighbors are communicating, potential next receiver TA > C+R+T C – contention interval length; R – RTS packet length; T – turn-around time, time bet. end of RTS and start of CTS; TA = 1.5 * (C+R+T);

22. Feb 16 2004 MAC for Sensor Networks 22 T-MAC: Overhearing Avoidance ~= S-MAC But implemented as an option in T-MAC Node – goes to sleep after overhearing RTS/CTS of other nodes communication miss other RTS/CTS transmissions disturb the medium while waking up throughput decreases Overhearing avoidance should not used when maximum throughput is required

23. Feb 16 2004 MAC for Sensor Networks 23 T-MAC: Asymmetric Communication (1) Early-Sleeping Problem – in convergecast (A to D) C – may lose medium to B (RTS) or A (B’s CTS) C loses to B; D will hear CTS from C; C loses to A; D will hear nothing, since C is silent;

24. Feb 16 2004 MAC for Sensor Networks 24 T-MAC: Asymmetric Communication (2) Future RTS (FRTS) Let others know that it cannot access the medium; C – sends FRTS – has duration field; receiver of FRTS – schedule timer; FRTS might affect data; so, DATA postponed until FRTS is over; Prevent others from taking medium, send dummy DS packet;

25. Feb 16 2004 MAC for Sensor Networks 25 T-MAC: Asymmetric Communication (3) Full-Buffer Priority – suitable for unidirectional flows Buffer – almost full – prefer sending than receiving Receive RTS, send its own RTS back instead of CTS Higher chance of transmitting its own message, lesser probability of early-sleeping, limited form of flow control

26. Feb 16 2004 MAC for Sensor Networks 26

27. Feb 16 2004 MAC for Sensor Networks 27 Homogeneous Local Unicast

28. Feb 16 2004 MAC for Sensor Networks 28 Nodes to Sink Communication ~= Convergecast

29. Feb 16 2004 MAC for Sensor Networks 29 Early-Sleeping Problem & Solutions Performance

30. Feb 16 2004 MAC for Sensor Networks 30 Event-Based Local Unicast

31. Feb 16 2004 MAC for Sensor Networks 31 Event-Based Local Unicast, Convergecast

32. Feb 16 2004 MAC for Sensor Networks 32

33. Feb 16 2004 MAC for Sensor Networks 33 How SS-S-TDMA is different, other than the obvious? S-MAC and T-MAC provide energy efficiency using a scheduling approach (TDMA idea) Protocol overhead – sending periodic SYNC messages, periodic neighbor discovery messages, and also synchronization issues SS-S-TDMA [KA04] has the following features: Low protocol overhead (initial slot assignment using diffusion, synchronization) Guaranteed message delivery No idle-listening, collisions, overhearing No 802.11 style RTS/CTS/DATA/ACKs Self-Stabilizing

34. Feb 16 2004 MAC for Sensor Networks 34