WearDrive: Fast and Energy-Efficient Storage for Wearables

Jian Huang, Anirudh Badam, Ranveer Chandra, Edmund B. NightingaleUSENIX ATC 2015 · OpenReview

TL;DR

Wearables are power-constrained devices with limited storage. This paper introduces WearDrive, a novel storage system that leverages a nearby smartphone to offload energy-intensive storage tasks. By using battery-backed RAM (BB-RAM) on the wearable for fast, local operations and a hybrid Bluetooth/Wi-Fi communication scheme to transfer data to the phone for durable storage, WearDrive significantly improves performance and battery life for wearable applications.

What Problem Does This Paper Address?

Wearable devices, such as smartwatches and fitness trackers, face a fundamental tension between their desire for rich functionality and their severe limitations in battery capacity and local storage. The need for durable and secure storage for personal data (e.g., health metrics, location) exacerbates this problem, as these operations are typically energy-intensive. Traditional mobile storage solutions, which rely on flash memory, are slow and consume significant power, leading to frequent charging needs for wearables. Furthermore, the reuse of mobile operating system designs on wearables, which are not optimized for their unique constraints, amplifies these energy overheads. This paper aims to bridge this gap by proposing a storage system that intelligently offloads tasks to a more powerful, nearby device (a smartphone) while minimizing energy consumption and ensuring data durability.

Methods

WearDrive's core innovation lies in its distributed storage architecture and hybrid communication strategy:

  • Battery-Backed RAM (BB-RAM): WearDrive utilizes BB-RAM on the wearable as its primary storage. This is essentially DRAM that is kept powered by the device's battery, allowing it to retain data even during temporary power interruptions or OS crashes. This provides fast, low-latency access, crucial for interactive wearable applications.
  • Phone as Secondary Storage: The smartphone acts as the durable, secondary storage. Data written to the wearable's BB-RAM is asynchronously transferred to the phone. The phone then handles energy-intensive operations like encryption and writing to its own flash storage.
  • Hybrid Bluetooth/Wi-Fi (HYN) Communication: To efficiently transfer data between the wearable and the phone, WearDrive employs a hybrid approach. It uses Bluetooth Low Energy (BLE) for low-power, always-on connectivity and for small data transfers. For larger data transfers, it intelligently switches to Wi-Fi Direct (WFD) to leverage its higher bandwidth and lower energy cost per bit, avoiding the high energy overhead of establishing a WFD connection for every small transfer.
  • WearCache and WearKV: The system is structured into two components: WearCache on the wearable, which acts as a fast, local cache for recent data, and WearKV on the phone, which serves as the main durable storage. Both components expose a key-value store interface, familiar to developers.
  • Crash Recovery: WearDrive incorporates a firmware component to ensure data durability in BB-RAM even in the event of an OS crash. It also handles warm resets by preserving BB-RAM state.

Datasets

The paper does not explicitly mention the use of pre-existing datasets for training. Instead, it introduces WearBench, a framework designed to generate synthetic data and workloads representative of typical wearable applications. WearBench simulates scenarios like extended display (notifications, social media) and sensor data analysis (fitness tracking, health monitoring) by creating configurable data streams and inter-arrival times. This allows for reproducible evaluation of storage systems on wearables.

Key Insights and Ideas

  • Trading Phone Battery for Wearable Battery: The most significant insight is the explicit trade-off: WearDrive leverages the phone's larger battery to power energy-intensive storage operations, thereby extending the wearable's battery life. This is a practical approach given the significant battery capacity difference between typical wearables and smartphones.
  • BB-RAM as a Practical Non-Volatile Storage: By using battery-backed DRAM, WearDrive achieves the performance benefits of RAM while providing durability comparable to flash, but with significantly lower energy costs for writes. The key is that wearable batteries are often non-removable, making DRAM less volatile in practice.
  • Hybrid Networking is Crucial for Energy Efficiency: Simply using BLE or WFD exclusively is suboptimal. The hybrid approach, intelligently switching between BLE for small/infrequent transfers and WFD for larger/frequent ones, is critical for minimizing overall communication energy.
  • Offloading is More Than Just Storage: By offloading storage to the phone, WearDrive also implicitly enables more complex data analytics and computations to be performed on the phone, further reducing the computational burden on the power-constrained wearable.

Key Results

WearDrive demonstrates substantial improvements across various metrics:

  • Performance: WearDrive improves the performance of wearable applications by up to 8.85x compared to "WearableOnly" (local flash storage).
  • Energy Efficiency: WearDrive reduces energy consumption on the wearable by up to 3.69x compared to "WearableOnly" for storage operations.
  • Battery Life: WearDrive can improve wearable battery life by up to 3.69x in specific scenarios. For example, passive heart-rate monitoring sees a 39% improvement, and batched notifications see a 149% improvement compared to Google WearSDK.
  • Phone Impact: The energy impact on the phone is negligible, accounting for only 5.1% of its battery life for workloads that save 16% on the wearable.
  • Communication Efficiency: The hybrid HYN protocol offers significant energy savings, with WearDrive being up to 1.81x more energy efficient than WearSDK-HYN due to BB-RAM's fast durability.

Limitations and Future Work

  • BB-RAM Dependency: WearDrive's durability relies on the battery maintaining power to the BB-RAM. While the system has mechanisms for graceful shutdown and crash recovery, a complete battery drain without warning could still lead to data loss.
  • Flash Overhead on Phone: While WearDrive offloads energy-intensive writes to the phone, the phone still eventually writes to its own flash, which has its own energy costs.
  • Future Work: The authors suggest providing filesystem and database interfaces using BB-RAM, exploring offloading computation to the cloud, and further optimizing the hybrid networking protocol. They also mention the potential for WearDrive to enable new functionalities by allowing wearables to span data and computation across devices.

Summary updated

BibTeX
@inproceedings{huang2015weardrive,
  author = {Jian Huang and Anirudh Badam and Ranveer Chandra and Edmund B. Nightingale},
  title = {{WearDrive: Fast and Energy-Efficient Storage for Wearables}},
  booktitle = {USENIX ATC},
  year = {2015},
  url = {https://openreview.net/forum?id=vRalwLVOwP},
}
Figure