WiOpt 2013 Life-Add: A novel WiFi design with battery life, throughput and fairness improvement Shengbo Chen*, Tarun Bansal*, Yin Sun*, Prasun Sinha and Ness B. Shroff Dept. ECE & CSE, The Ohio State University Background Battery life is a serious problem for most smartphone users WiFi, 4G LTE, GPS, Bluetooth, screen, CPU, ... Web browsing via WiFi Test results in April 2013 by Battery life < 11 hours for most popular smartphones iPhone 5 802.11n HTC One 802.11ac Samsung Galaxy S 4 802.11ac 2 Existing Solutions to Prolong Lifetime Mobile Charging Additional equipment Solar charger portable battery wireless charger Reduce power when sensing Lower hardware clock-rate [E-MiLi, Mobicom 11] Broadcom SoC Solution —802.11 ac —Used in HTC One and Samsung Galaxy S 4 —Test: 7.8 hours by Trade bandwidth/throughput for power reduction —Cannot have both benefits 3 IEEE 802.11 Standard Evolution WLAN 802.111997 2 Mbps, DSSS, FHSS 802.11b 11 Mbps, CCK, DSSS 802.11a 54 Mbps, OFDM, 5 GHz 802.11g 54 Mbps, OFDM, 2.4 GHz 802.11n 600 Mbps with 4x4 MIMO, 20/40 MHz BW, 2.4 or 5 GHz 802.11ac 256QAM 160MHz 802.11ad 802.11p 27 Mbps,10 MHz BW, 5.9 GHz Wireless Access for Vehicular Environment 802.11af TVWS TV White Spaces Wireless Gigabit, <6 GHz Wireless Gigabit, 60 GHz Physical layer Significant evolutions towards high throughput MAC CSMA/CA and its enhancements — QoS, security, frame aggregation, block ACK 4 Can we do better? Life-Add: An innovative MAC design Battery Lifetime —Avoid unnecessary sensing Throughput —Reduce collisions and starvations Fairness —Near-far effect B L I F E T I M E N F A I R N E S S F I T H R O U G H P U T 5 Contents Background Life-Add: An innovative MAC design Simulation Results Summary 6 Life-Add: Smartphone energy model Power source: Strong: Wall power, portable battery Weak: Solar charger Other components 4G LTE, CPU, screen, … WiFi chip ON: Transmit/receive/sensing —High power consumption OFF: Sleep —Very low power consumption Too much sensing means a significant waste of energy Sleep/wake (asynchronous) 7 Life-Add: Sleep/Wake + Channel Contention Uplink Device 1 Device 2 ACK AP 8 Life-Add: Sleep/Wake + Channel Contention Uplink Device 1 wakes up earlier and senses the channel Device 1 Device 2 ACK AP 9 Life-Add: Sleep/Wake + Channel Contention Uplink Device 1 transmits, Device 2 goes back to sleep Data Device 1 Device 2 ACK AP 10 Life-Add: Sleep/Wake + Channel Contention Uplink AP replies an ACK to Device 1. Cycle 1 completes. Data Device 1 Device 2 ACK ACK AP Cycle 1 11 Life-Add: Sleep/Wake + Channel Contention Uplink Devices 1 and 2 wake up at almost the same time Data Device 1 Device 2 ACK ACK AP Cycle 1 12 Life-Add: Sleep/Wake + Channel Contention Uplink A collision occurs, followed by a timeout. Cycle 2 completes. Data Data Device 1 Data Device 2 ACK ACK AP Cycle 1 Cycle 2 13 Life-Add: Sleep/Wake + Channel Contention Uplink Data Data Device 1 Data Data Device 2 ACK ACK AP Cycle 1 Cycle 2 A new renewal process model: each cycle is an i.i.d. period —Requires 2 assumptions: — Exponential distributed sleep period: from last cycle) Memoryless (independent —Tdata + TACK≈ Tcollision + Ttimeout (only assumed in analysis, not in simulations) 14 Life-Add vs IEEE 802.11 Life-Add Data Data Device 1 Data Data Device 2 ACK ACK AP Cycle 2 Cycle 1 IEEE 802.11 Data Data Device 1 Data Data Device 2 ACK ACK AP Sleep backoff vs sensing backoff (save energy) Renewal process vs 2D Markov chain [Bianchi 2000] (simplify optimization) 15 Life-Add: Downlink A short Ps-poll packet is used to contend for the channel Ps-poll Ps-poll ACK Device 1 ACK Device 2 Ps-poll Data Data AP Beacon Cycle 1 Cycle 2 Still a renewal process Uplink: sleep + data + overhead (ACK/collision/timeout) Downlink: sleep + data + overhead (ACK/Ps-poll/collision/timeout) —Additional Ps-poll packet as part of overhead Can be modeled together 16 Life-Add: Single AP Proportional-fair Utility Maximization max ∑ log E{Throughput of Device i} s.t. E{Battery Life of Device i} ≥ Tmin,i 17 Life-Add: Single AP Proportional-fair Utility Maximization max ∑ log E{Throughput of Device i} s.t. Pr{Device i’s RF is ON}≤ bi Maximal device-ON probability: bi Variables: average sleep period 1/Ri 18 Life-Add: Single AP Proportional-fair Utility Maximization max ∑ log E{Throughput of Device i} s.t. Pr{Device i’s RF is ON}≤ bi Maximal device-ON probability: bi Variables: average sleep period 1/Ri Non-convex —Asynchronous network with collisions —Channel access probabilities of the devices are coupled We propose a solution: Life-Add Theorem: Asymptotically optimal, as Tsensing /(Tdata + TACK)0 — E.g., 802.11b: Tsensing= 4us, Tdata + TACK=511us~1573us 19 Problem formulation where , is a scaling constant is the transmission success probability is the device-ON probability Proof idea: Problem structure, KKT necessary conditions Upper and lower bounds converge to the same value 20 Life-Add: Single AP Pr{Device i’s RF is ON}≤ bi Implementation procedure: Each device reports bi to the AP The AP computes —If , —If , Device n uses Use and , and broadcast them to the devices to compute to generate the sleeping period Low complexity, easy to implement 21 Life-Add: Single AP NS-3 simulation for a homogeneous scenario Red curve: simulated performance with no approximation Blue point: closed form solution of Life-Add Observation: Life-Add is near optimal The renewal process model is reasonably accurate 22 Life-Add: general multiple APs Too complicated interference model Global optimization is very difficult Near-far effect Device 1 can access the channel all the time Device 2 is in starvation Hidden terminal problem Two devices cannot sense each other and cause collisions 23 Life-Add: general multiple APs Near-far effect Node collaboration —Device 1 computes the two values of average sleep period suggested by AP 1 and AP 2 —Device 1 chooses the longest average sleep period to reduce collisions with Device 2, which is vulnerable To care for the vulnerable 24 Life-Add: general multiple APs Hidden terminal problem Increase average sleep period after a collision Reset average sleep period after a successful transmission —Similar idea to 802.11 MAC 25 Life-Add: general multiple APs Implementation procedure: Each device reports bi to nearby APs Each AP computes and broadcasts —If , —If , Device n uses and to compute Choose to use the smallest Reduce Use and at collision, reset suggested by nearby APs value after receiving ACK to generate the sleeping period 26 Life-Add: general multiple APs NS-3 simulation results: Uplink: 4 APs, 30 smartphones, randomly located in a 500×500 m field, UDP saturation bi = 1 no lifetime (power-ON prob.) constraints 1/3 with battery, 1/3 with battery + solar panel, 1/3 to wall power Battery level: uniform distribution within 200~1000 mAh Lifetime and throughput benefits 27 Life-Add: general multiple APs NS-3 simulation results: Per-device performance: Battery life improvement for all 5 devices Significant throughput increase for the low-rate device 28 Life-Add: general multiple APs Average performance gain Battery Life: —Sleep/Wake Throughput: —Node collaboration (reduce collisions and starvations) —Parameter optimization Fairness: —Node collaboration (to care for the vulnerable) —Proportional-fair utility 29 Life-Add: general multiple APs Coexisting with IEEE 802.11 AP 1,2 and their users upgrade from IEEE 802.11 to Life-Add Battery life — Longer if you use Life-Add Throughput — Higher no matter you use Life-Add or not, due to less collisions 30 Summary A novel renewal process model for energy efficient WiFi design Proportional-fair utility maximization problem Non-convex Life-Add MAC design Near optimal for single AP cases Alleviate “near-far effect” and “hidden terminal problem” in general cases Easy to implement Ns-3 simulations Battery life, throughput, and fairness improvement Coexists harmoniously with IEEE 802.11 Not just WiFi: Last-hop decentralized access Internet of Things, Military,… US patent filed 31 Tasks to do… More simulations for joint uplink and downlink Practical traffics Web browsing, video streaming, email, searching Hardware testing 32 B L I F E T I M E N F A I R N E S S F I T H R O U G H P U T 33