Snooze: Energy Management in n WLANs Ki-Young Jang, Shuai Hao, Anmol Sheth, Ramesh Govindan
Background 2 Evolution of Wi-Fi Performance vs. Energy Usage 8x speedup compared to a/g Mbps
Background n Features and Energy Usage A A B B Energy management should exploit both sleep opportunities and antenna configuration. Energy management should exploit both sleep opportunities and antenna configuration. MIMO higher data rate spatial diversity Additional power states: 10-30% of peak power consumption of the tablet! Additional power states: 10-30% of peak power consumption of the tablet! # of Antenna Intel Wi-Fi Link 5300(W) TxRxIdleSleep x and 1.6x 1.3x and 1.7x
MotivationDesignImplementationEvaluation Motivation 4 Micro-sleep Opportunities A B AP t Sleep! Z Z Z Z Z Z t t
Micro-sleep Sleep + Wakeup = ~2ms MotivationDesignImplementationEvaluation Motivation 5 Micro-sleep Opportunities # of STA Traffic Sleeping while AP is servicing others 1 1 Sleeping during inter-frame gaps 2 2 Depending on the traffic and number of clients, we can get energy savings of 30% ~ 90%.
MotivationDesignImplementationEvaluation Motivation 6 Internet as bottleneck: SISO is best Internet as bottleneck: SISO is best High bandwidth scenario: MIMO3 is best Antenna Configuration Antenna configuration should be adaptive based on traffic demand and link quality. Antenna configuration should be adaptive based on traffic demand and link quality.
Challenges 7 Micro-sleep with minimal impact on delay and throughput-sensitive traffic Adaptive antenna configuration management Joint design of both mechanisms Application agnostic Time (ms) Power(W) Sleep for 50ms Sleep for 70ms
Contribution 8 Design and Implementation of Snooze Joint, application-agnostic design of client micro- sleep and antenna configuration management. Extensive experiments that demonstrate 30~85% energy-savings over CAM across a wide range of traffic scenarios.
MotivationDesignImplementationEvaluation AP-Directed Design 9 Snooze AP Snooze Client Shapes traffic to create sleep opportunities Minimal impact on traffic Minimizes the number of active clients Manages antenna configurations Minimizes antennas needed Goal: Reduce client energy consumption by jointly controlling sleep and antenna configuration
Sleep duration: based on measured packet arrival rate Awake duration: based on average airtime consumption Sleep duration: based on measured packet arrival rate Awake duration: based on average airtime consumption MotivationDesignImplementationEvaluation Snooze Components 10 Micro-sleep Scheduling Micro-sleep Scheduling Antenna Mgmt. Antenna Mgmt. If measured airtime utilization is turn off 1 antenna > 0.7: low link quality or less antennas -> turn on 1 antenna Rate Adaptation: AP uses default rate-control algorithm with restricted search space If measured airtime utilization is turn off 1 antenna > 0.7: low link quality or less antennas -> turn on 1 antenna Rate Adaptation: AP uses default rate-control algorithm with restricted search space Hysteresis and moving averages A: 1Mbps B: 20Mbps Time (Second)
MotivationDesignImplementationEvaluation Implementation APClient 11 Driver Kernel Rate Table Rate Table Airtime Usage Airtime Usage iwlagn Airtime Scheduling Sleep/Wakeup Computation Sleep/Wakeup Computation mac Per client traffic queue Sleep/ Wakeup Sleep/ Wakeup Antenna Configuration iwlagn mac80211
MotivationDesignImplementationEvaluation Evaluation 12 Applications used for evaluation Delay/Jitter Sensitivity HighLow Bandwidth Requirement HighHD video streamingFile downloading LowVoIPChat Overview and Setup Performance comparison Constantly Awake Mode (CAM) Power Save Mode (PSM) Evaluation metric Total energy usage of NIC Application throughput and delay
MotivationDesignImplementationEvaluation Evaluation 13 High Definition Video Streaming More than 25% low-power sleeping compared with both CAM and PSM More than 25% low-power sleeping compared with both CAM and PSM About 50% energy savings compared with both CAM and PSM About 50% energy savings compared with both CAM and PSM Average delay CAM : 2.5ms, PSM : 4ms, Snooze : 8ms Average delay CAM : 2.5ms, PSM : 4ms, Snooze : 8ms
MotivationDesignImplementationEvaluation Evaluation 14 Heterogeneous Traffic Snooze can accommodate multiple concurrent applications. Client Mode One app per client Both techniques contribute significantly to energy savings, but contribution varies across traffic. Energy saving breakdown File VoIP HD Chat micro- sleep antenna config.
Related Work 15 Energy Management Techniques for uPMC-PSMCatnapNAPmanSnooze AP-directed Traffic types HB-DS HB-DI LB-DS LB-DI n Rate adaptation Multiple apps per client
Conclusion 16 Client micro-sleep and antenna configuration management Application agnostic 30~85% energy saving across a wide range of traffic scenarios Snooze: Energy Management Scheme for n Future Work Multi-AP setting Highly bursty workloads Parameter sensitivity