WearDrive: Fast and Energy Efficient Storage for Wearables

Transcription

1 WearDrive: Fast and Energy Efficient Storage for Wearables Reza Shisheie Cleveland State University CIS 601

2 Wearable Computing: A New Era 2

3 Wearable Computing: A New Era Notifications Fitness/Healthcare Audio/Video Data Analytics 2

4 Wearables Are Battery Constrained K mah 3

5 Wearables Are Battery Constrained K mah f (Size & Weight) mah 3

6 Wearables Are Battery Constrained K mah f (Size & Weight) mah 3

7 Wearables Are Battery Constrained K mah f (Size & Weight) mah 3

8 Wearables Are Battery Constrained K mah Mobile SoC, OS & SDK f (Size & Weight) mah Reduce Development Cost Wearable SoC, OS & WearSDK 3

9 Wearables Are Battery Constrained K mah Mobile SoC, OS & SDK f (Size & Weight) mah Reduce Development Cost Wearable SoC, OS & WearSDK Offload Wearable s Workload to the Phone 3

10 Energy Challenge: Wearable Apps Are Data-Intensive Extended Display Notifications 4

11 Energy Challenge: Wearable Apps Are Data-Intensive Extended Display Sensors Notifications 16+ Sensors 4

12 Energy Challenge: Wearable Apps Are Data-Intensive Extended Display Sensors Fast & Energy-Efficient Storage System Notifications 16+ Sensors 4

13 Mobile Storage is Energy-Intensive Flash device is not energy-intensive, Storage software stack is energy-intensive! 5

14 Energy Overhead Factors Mobile Storage is Energy-Intensive Flash device is not energy-intensive, Storage software stack is energy-intensive! KB 32 KB 0 Random Read Random Write Sequential Read Sequential Write Storage software consumes x more energy than Flash [Li et al., FAST 14] 5

15 Mobile Storage is Energy-Intensive Flash device is not energy-intensive, Storage software stack is energy-intensive! + + Slow Flash: Increased CPU Idle Time Runtime System Overhead Data Encryption 5

16 Use DRAM as Storage? Applications DRAM OS Flash DRAM is fast Closer to applications No software overhead for isolation and security 6

17 Use DRAM as Storage? Applications DRAM OS Flash DRAM is fast Closer to applications No software overhead for isolation and security DRAM is volatile, the data durability is a problem! 6

18 Leveraging Batteries for Non-Volatility DRAM Battery = Non-Volatile Memory 7

19 Leveraging Batteries for Non-Volatility DRAM Battery = BB-RAM Non-Volatile Memory 7

20 Leveraging Batteries for Non-Volatility DRAM Battery = BB-RAM Non-Volatile Memory No Hardware Changes 7

21 Leveraging Batteries for Non-Volatility DRAM Battery = BB-RAM Non-Volatile Memory + No Hardware Changes Performance Improvement 7

22 Leveraging Batteries for Non-Volatility DRAM Battery = BB-RAM Non-Volatile Memory + + No Hardware Changes Performance Improvement Data Durability 7

23 Our BB-RAM Implementation BB-RAM page DRAM page 8

24 Our BB-RAM Implementation BB-RAM page DRAM page BB-RAM Manager 8

25 Our BB-RAM Implementation BB-RAM page DRAM page Memory Pressure BB-RAM Manager Automatically adjust BB-RAM size according to application requirement 8

26 Our BB-RAM Implementation OS Crashes BB-RAM page DRAM page BB-RAM Manager 8

27 Our BB-RAM Implementation BB-RAM page DRAM page OS Deadlock BB-RAM Manager A watchdog timer to detect if OS is hung 8

28 Leveraging Phone for Capacity/Compute Wearable applications focus on the latest data, but how about the old data? 9

29 mj/kb Leveraging Phone for Capacity/Compute Wearable applications focus on the latest data, but how about the old data? Local Flash Remote Memory via WiFi Direct IO Size (KB) Two Moto E phones with b/g/n Writing data to remote memory via WFD is more energy efficient 9

30 Reaching the Phone Efficiently Standby Power Data Transfer BLE Low 0.41 microjoules/bit WFD High 0.02 microjoules/bit BLE for small data transfer, WFD for large data transfer BLE WiFi Direct 10

31 Reaching the Phone Efficiently Standby Power Data Transfer BLE Low 0.41 microjoules/bit WFD High 0.02 microjoules/bit BLE for small data transfer, WFD for large data transfer BLE WiFi Direct Use BLE s connection quality for reference 10

32 Building A Distributed In-Memory Storage System Application BB-RAM Cache Wearable Efficient Wireless Network BB-RAM + Flash Storage Phone 1 Improve performance & save energy 2 Reduce Flash size reduce $ cost (10%) 3 Leverage phone s resources for wearable 11

33 WearDrive Design: Key Value Store Application BB-RAM WearCache Wearable Efficient Wireless Network BB-RAM + Flash WearKV Phone 12

34 WearDrive Design: Key Value Store Application BB-RAM WearCache Wearable Efficient Wireless Network BB-RAM + Flash WearKV Phone + + F I F O Why Key Value Store? Focus on Latest Data Hybrid Networks 12

35 Putting It All Together WearCache WearKV Wearable Hybrid BLE/WFD Phone 13

36 Putting It All Together WearCache WearKV Wearable Hybrid BLE/WFD Phone 13

37 Putting It All Together WearCache WearKV Wearable Hybrid BLE/WFD <key, value> Phone 13

38 Putting It All Together WearCache Notification WearKV Wearable Hybrid BLE/WFD <key, value> Phone 13

39 Putting It All Together WearCache Notification WearKV <key, value> Wearable Hybrid BLE/WFD <key, value> Phone 13

40 Putting It All Together WearCache Notification WearKV <key, value> Wearable Hybrid BLE/WFD <key, value> Phone 13

41 Putting It All Together WearCache WearKV Wearable Hybrid BLE/WFD Phone 13

42 Putting It All Together RegisterForSensor (Sensor ID) WearCache WearKV Wearable Hybrid BLE/WFD Phone 13

43 Putting It All Together RegisterForSensor (Sensor ID) WearCache <key, value> Wearable Hybrid BLE/WFD WearKV Phone 13

44 Putting It All Together RegisterForSensor (Sensor ID) WearCache WearKV <key, value> Wearable Hybrid BLE/WFD <key, value> Phone 13

45 Putting It All Together RegisterForSensor (Sensor ID) WearCache WearKV <key, value> Wearable Hybrid BLE/WFD <key, value> Phone RegisterCallBack (TimeGap, CallBackFun) 13

46 Putting It All Together RegisterForSensor (Sensor ID) WearCache WearKV <key, value> Wearable Hybrid BLE/WFD <key, value> Phone RegisterCallBack (TimeGap, CallBackFun) 13

47 Putting It All Together WearCache <key, value> RegisterForSensor (Sensor ID) Wearable Notification Hybrid BLE/WFD WearKV <key, value> Phone RegisterCallBack (TimeGap, CallBackFun) 13

48 WearDrive: Implementation in Real System Application Android OS (version 4.4) Mobile Device 14

49 WearDrive: Implementation in Real System Application KV-Store: WearCache/WearKV BB-RAM Manager Hybrid Network Management Android OS (version 4.4) Mobile Device 14

50 WearDrive: Implementation in Real System Application WearDrive API KV-Store: WearCache/WearKV BB-RAM Manager Hybrid Network Management Android OS (version 4.4) Mobile Device 14

51 Experimental Setup Emulated wearable device (similar to Samsung Gear) 1.2 GHz Dual-core MB RAM + BLE WiFi b/g/n Monsoon power monitor 15

52 Experimental Setup WearableOnly WearSDK WearDrive Use local Flash as storage Android Wear SDK In-memory storage system Its data layer was expanded to make it support WFD and hybrid network protocol. 15

53 WearBench: Benchmarks for Wearables Examples Parameters Extended Display Sensors Audio/Video Size, inter-arrival time Sampling rate, monitoring period Encoding rate, quality, monitoring period 16

54 Energy (Joules) Throughput (K TPS) Local Memory vs. Local Flash (Write) WearableOnly 2 4 threads WearDrive IO Size (Bytes) x performance improvement, 4x less power 17

55 Energy (Joules) Throughput (K TPS) Local Memory vs. Local Flash (Read) WearableOnly 4 threads WearDrive IO Size (Bytes) x performance improvement, 1.7x less power 18

56 Energy (Joules) Energy Usage of Sensor Data Aggregation Storage Networking WearableOnly WearSDK-BLE WearSDK-WFD WearSDK-HYN WearDrive The sensor samples values continuously at 1 Hz for 24 hours WearDrive consumes 1.5x less power than WearableOnly, up to 3x less power than WearSDK 19

57 Energy (Joules) Energy Usage of Receiving Notifications 12 WearSDK-BLE WearSDK-WFD WearSDK-HYN WearDrive Interval between two notifications (seconds) WearDrive saves 1.2-3x energy than WearSDK-BLE 20

58 Compute Offload & Impact on the Phone Wearable (300 mah) Phone (2000 mah) WearableOnly 27.12% -- WearableOnly +InMemory 13.23% -- WearDrive 0.87% 2.09% Data analysis with k-means on the data set generated by 16 sensors at 1 Hz for 24 hours. 21

59 Conclusion WearDrive: Storage System for Wearables In-memory Storage System with battery-back DRAM (BB-RAM) Extended Capabilities leveraging phone s storage and CPU Extended Lifetime 3x improvement on battery life time 22

60 Thanks! Q&A