Joe Polastre B-MAC Factored MAC protocol exposing control of sub-primitives

2 Outline Factored MAC protocol design  Information sharing with higher layers  Control and reconfiguration of link protocol  Tradeoffs in link protocols Feedback mechanism Link Abstraction for Sensor Networks

3 B-MAC Design Principles  Reconfigurable MAC protocol  Flexible control  Hooks for sub-primitives Backoff/Timeouts Duty Cycle Acknowledgements  Feedback to higher protocols  Minimal implementation  Minimal state Primary Goals  Low Power Operation  Effective Collision Avoidance  Simple/Predicable Operation  Small Code Size  Tolerant to Changing RF/Networking Conditions  Scalable to Large Number of Nodes Our implementation is on Mica2 motes with CC1000

4 B-MAC Link Protocol Interaction Reconfiguration and control of link layer protocol parameters  Acknowledgements, Backoff/Timeouts, Power Management, Hidden Terminal Management Ability to choose tradeoffs – “knobs”  Fairness, Latency, Energy Consumption, Reliability Power consumption estimation through analytical and empirical models  Feedback to network protocols  Lifetime estimation Mechanisms to achieve network protocols’ goals

5 S-MAC Ye, Heidemann, and Estrin, INFOCOM 2002 Traditional monolithic protocol design Synchronized protocol with periodic listen periods “Black Box” design  Designed for a general set of workloads  User sets radio duty cycle  SMAC takes care of the rest so you don’t have to  Integrates higher layer functionality into link protocol T-MAC [van Dam and Langendoen, Sensys 2003]  Reduces power consumption by returning to sleep if no traffic is detected at the beginning of a listen period Schedule 2 Schedule 1 Wei Ye, USC/ISI Node 1 Node 2 sleep listen sleep listen sleep sync

6 Low Power Listening (LPL) Higher level communication scheduling  Energy Cost = RX + TX + Listen  Start by minimizing the listen cost Example of a typical low level protocol mechanism Periodically  wake up, sample channel, sleep Properties  Wakeup time fixed  “Check Time” between wakeups variable  Preamble length matches wakeup interval Overhear all data packets in cell  Duty cycle depends on number of neighbors and cell traffic RX wakeup TX sleep Node 2 Node 1 time

7 Effect of LPL Check Interval Single hop data reporting application Higher sampling rate  Higher traffic in a cell  Higher duty cycle Optimize the check time to the traffic  Application knows sample rate (packet generation rate)

8 Effect of Neighborhood Size Neighborhood Size affects amount of traffic in a cell  Network protocols typically keep track of neighborhood size  Bigger Neighborhood  More traffic Automatic Optimization  Mac protocol can monitor channel  Application vs network protocol vs link protocol

9 Factored vs Layered Protocols Experimental Setup:  n nodes send as quickly as possible to saturate the channel Factored link layer never worse than traditional approach  Often much better Flexible configuration yields efficient:  Reliable transport (Acks)  Hidden Terminal support (RTS-CTS) ProtocolROMRAM B-MAC B-MAC w/ ACK B-MAC w/ Duty Cycling B-MAC w/ DC & ACK S-MAC topology

10 Fragmentation Support Factored vs Layered Protocol S-MAC  RTS-CTS Fragmentation Support B-MAC  Network protocol sends initial data packet with number of fragments pending  Disable backoff & LPL for rest of fragments Measure energy consumption at C (bottleneck node) Minimizing power relies on controlling link layer primitives A B C E D Sometimes the black box is worse than the naïve approach

11 Tradeoffs: Latency for Energy Factored vs Traditional Protocol Assume a multihop packet is generated every 10 sec  No queuing delay allowed Delay the packet  S-MAC sleeps longer between listen period  B-MAC increases the check interval and preamble length S-MAC Default Configuration B-MAC Default Configuration

12 Tradeoffs: Throughput for Energy Factored vs Layered Protocol 10 node single hop network  Increase transmission rate  Deliver each packet within 10 sec  Measure average power consumption per node As throughput increases  B-MAC reduces check interval as traffic increases  S-MAC uses optimal duty cycle Protocol overhead causes energy to increase linearly topology

13 Surge Application Run B-MAC in a real world application  8 days/40000 data readings in deployment Surge Multihop Data Collection includes:  Data reporting every 3 minutes  B-MAC check:sleep ratio: 1:100  Jason Hill’s ReliableRoute Reliability improvements to MintRoute Turn on link layer acks in B-MAC Add retransmissions  Power metering in the link protocol Simple routing protocol optimization  Turn off long preambles when sending to the base station (one hop away)  Base station always on Network configuration images from Jason Hill

14 Surge Application Network power consumption of a factored link protocol Duty cycle dependant on position in network  Leaf nodes have least amount of traffic  Middle nodes forward leaf node traffic  Nodes 1 hop from base station benefit from reconfiguration 2.35% worst case node duty cycle Leaf Nodes 1 hop from base station Forwarded ~35,000 (85%) packets Duty cycle 75% higher without optimization Forwarded <10,000 packets

15 Tradeoffs: Latency vs Reliability Surge Application Reliability  98.5% of all packets delivered  Some nodes achieved an astounding 100% delivery …but communication links are volatile  Retransmissions required  After n retries, give up and pick a new parent Actual latency  Follows expected latency from microbenchmark  Retransmission delay  Contention delay (infrequent)

16 Lifetime Model B-MAC Analytical Model  Worst case analysis  Calculate Optimum LPL parameters Preamble length  Evaluate Neighborhood size Sample rate  Minimize Power consumption  Verify Experimental operation matches analytical calculations  Runtime Feedback mechanism Effect of turning knobs  t i : LPL check interval  r : Sample rate (packet generation rate) Knowledge from network protocols  n : Neighborhood size Determine  t : Fraction of each second spent on each operation  E : Energy consumed by each operation per second

17 Lifetime Model Transmit Receive NotationParameter rSample Rate (packets/sec) nNeighborhood size L preamble Preamble length (bytes) L packet Packet length (bytes) c sleep Current : Sleep (mA) c rxb Current : Rx one byte c txb Current : Tx one byte C batt Capacity : Battery (mAh) VVoltage titi Time : Radio sampling interval (s) t startup Time : Radio startup t rxb Time : Rx one byte t rx Time : Rx per second t txb Time : Tx one byte t tx Time : Tx per second tltl Time : Lifetime (s)

18 Lifetime Model LPL Sampling Sleep NotationParameter rSample Rate (packets/sec) nNeighborhood size L preamble Preamble length (bytes) L packet Packet length (bytes) c sleep Current : Sleep (mA) c rxb Current : Rx one byte c txb Current : Tx one byte C batt Capacity : Battery (mAh) VVoltage titi Time : Radio sampling interval (s) t startup Time : Radio startup t rxb Time : Rx one byte t rx Time : Rx per second t txb Time : Tx one byte t tx Time : Tx per second tltl Time : Lifetime (s)

19 Lifetime Model The total energy, E, can be used to calculate the expected lifetime of the system NotationParameter rSample Rate (packets/sec) nNeighborhood size L preamble Preamble length (bytes) L packet Packet length (bytes) c sleep Current : Sleep (mA) c rxb Current : Rx one byte c txb Current : Tx one byte C batt Capacity : Battery (mAh) VVoltage titi Time : Radio sampling interval (s) t startup Time : Radio startup t rxb Time : Rx one byte t rx Time : Rx per second t txb Time : Tx one byte t tx Time : Tx per second tltl Time : Lifetime (s)

20 Link Abstraction in Wireless Sensor Networks What have we learned from factored protocols?  Integration of routing and organization protocols with link protocols not necessary Why do current WSN protocols integrate?  IP Nets: Transport protocols set fragmenting, addressing type of service, retransmission etc  WSNs: Network protocols setting these parameters Each hop has lots of volatility – keep state at every hop Local per-link decisions to save power  IP model forces policy and mechanism to be integrated  Negates nice properties of IP abstraction: No protocol interchangeability, inefficient large implementations A new abstraction must describe what the system is doing at each link  Dynamically change link protocol mechanism  Meeting point between link and routing layers  Separates link mechanisms from routing/network policies Aggregation Routing Services “policy” Link Abstraction MAC Protocols Physical Layers “mechanism” Physical Medium Applications

21 Link Abstraction B-MAC shows the need for bidirectional information flow with network protocols  Exposing control is critical for long lived operation  Enable link protocol interchangeability underneath optimized network protocols (routing, aggregation, organization, etc)  Smallest, most powerful primitives to execute higher level protocols efficiently Proposed abstraction includes 3 things:  Control of link layer protocol parameters  Ability to choose tradeoffs – “knobs”  Power consumption feedback model Next steps:  An RFC describing the abstraction in detail (this summer)  Implementation and use of abstraction in TinyOS with B-MAC/ Aggregation Routing Services “policy” Link Abstraction MAC Protocols Physical Layers “mechanism” Physical Medium Applications

22 Conclusions Coordination with higher protocols is essential for long lived operation Feedback allows protocols to make informed decisions Monolithic protocols can benefit from factoring— but reconfiguration may prove costly Traditional abstraction at the network layer doesn’t fit sensor networks—need a new abstraction at the link layer

Backup Slides

24 WooMAC Woo and Culler, Mobicom 2001 Move protocol knowledge into the MAC implementation ARC Protocol  Change the rate of originating traffic  Allows route-through traffic to reach its destination  Tradeoff everything else for fairness (very effective) Adaptive Backoff  Change CSMA backoff depending on type of traffic Broadcast Multihop

25 UNPF Ding, Silvalingam, Kashyapa, and Chuan, Vehicular Technology Conference, 2003 Unified Network Protocol Framework (UNPF)  Network organization protocol,  Medium access control (MAC) protocol, and  Routing protocol. Integrated network and link layers Write new UNPF implementations for each application -or- Trust the larger “black box” RoutingOrganization Link Protocol Application PHY Abstraction UNPF Routing, Organization, and MAC

26 Other “Black Box” Protocols Energy-Aware TDMA-Based MAC  Arisha, Youssef, Younis, IMPACCT 2002  Gateway node centrally manages sensor network clusters  integrate organization protocol TRAMA: Collision Free Protocol  Rajendran, Obraczka, Garcia-Luna-Aceves, SenSys 2003  Keep track of 2-hop neighbors, schedule traffic, eliminate collisions  Adaptively change schedule as “inter-arrival” message time changes  integrate organization protocol, assume traffic pattern Sift: Event Driven MAC Protocol  Jamieson, Balakrishnan, Tay, MIT TR 2003  Small and fixed contention window to save energy  Probabilistically pick an empty slot Moral: encourage protocol designers to expose configuration and tradeoffs (Sift even has a power consumption model)

27 Tradeoffs: Latency vs Reliability Factored vs Traditional Protocol Additional protocol overhead adds latency  Choose applicable overhead  ACK  RTS-CTS (S-MAC)  Synchronization B-MAC  If multihop packet Wait >= 2 packet times to send next message Implicit reservation and hidden terminal support S-MAC  Reserves channel at each hop

28 Duty Cycle Calculation Check time is 2.5ms If check interval is 250ms, duty cycle is 1% right?  Wrong!  Duty cycle is amortized cost of waking up energy consumption to sleep energy consumption Breakdown:  Full power (e) 250ms  Half power (b,d,f) 600ms  Low power (c) 1500ms Equivalent to being at full power for 800ms

29 IEEE “Logical Link Control” IEEE Standard, 1998 Data transfer interface Establishes connections “Control” path ends at  Doesn’t pass down to underlying link protocols  Doesn’t expose underlying capabilities of 802.3, , etc Instance specific code above Linux implementation  Used as a programming interface, not an abstraction  Heavyweight: 9008 lines of source, 28kB compiled Not an effective abstraction and not flexible Bridging / Convergence Logical Link Control Service Provider (802.2 LLC) Request Indication Response Confirm Service User

30 IEEE “Low Rate Wireless Personal Area Networks” IEEE Standard, 2003 Designed for low power, low data, high density rate wireless networks Services may interface with either or directly with the MAC protocol has lots of extra functionality that may never be used:  Enable the development of a “lightweight MAC”  Use 15.4 hardware without using the 15.4 MAC protocol (eg: B-MAC, S-MAC, etc) MAC PHY “Upper Layers” WSN Data Link Layer Abstraction Other MAC PHY x