TL;DR
This paper investigates whether offloading communication-intensive tasks from mobile apps to the cloud can save battery life. They found that while offloading can save energy, it's often hindered by app constraints and requires careful analysis to identify profitable opportunities, leading them to develop tools for constraint detection and energy profiling.
What Problem Does This Paper Address?
Mobile devices are increasingly powerful, but their limited battery life remains a significant bottleneck. Offloading computation to the cloud has been explored as a solution to extend battery life. However, many popular mobile applications, like social media clients and email apps, spend a substantial amount of energy on network communication rather than heavy computation. This paper addresses the question: can offloading these communication-intensive tasks to the cloud also lead to significant energy savings? The authors highlight that existing research has largely overlooked this aspect, focusing primarily on computational offloading. They aim to bridge this gap by exploring the feasibility and challenges of offloading network communication in real-world applications.
Methods
The authors employ a two-pronged approach:
- Case Study with AndTweet: They use an open-source Twitter client, AndTweet, as a testbed to explore offloading communication-related methods. They leverage the ThinkAir framework, which allows for method-level offloading to a remote surrogate machine.
- Constraint Identification Tool: To automate the detection of issues that prevent offloading, they develop a static source code analysis tool. This tool identifies constraints related to hardware access (e.g., device-specific APIs) and software limitations (e.g., non-serializable classes, state synchronization issues) that are common in Android applications.
- Communication Energy Profiling Tool: To accurately estimate energy savings, they create a tool that profiles communication energy consumption at a per-method level. This tool correlates network traffic traces with program execution traces, using power models that account for factors like traffic patterns and "tail energy" (energy wasted when the radio stays on after transmission).
Datasets
The paper does not introduce new datasets. Instead, it uses:
- Open-source Android applications: Specifically, AndTweet (a Twitter client) and K-9 Mail (an email client) are used as case studies.
- A set of 16 open-source applications: These are analyzed by the constraint identification tool to gather statistics on offloading feasibility.
Key Insights and Ideas
- Communication offloading can save energy: The study demonstrates that offloading network communication tasks can indeed lead to energy savings, particularly in Wi-Fi environments.
- App constraints are a major hurdle: Real-world applications are not designed for offloading. They contain numerous constraints (e.g., reliance on device-specific hardware, non-serializable objects, state synchronization issues) that make direct offloading difficult or impossible without code modifications.
- Manual constraint identification is tedious: Identifying these constraints manually is a time-consuming and error-prone process, highlighting the need for automated tools.
- Energy savings are highly context-dependent: The profitability of offloading is not guaranteed and depends heavily on factors like network conditions (Wi-Fi vs. 3G), traffic patterns, and latency. Offloading can even increase energy consumption in certain scenarios (e.g., high latency 3G networks).
- Asynchronous APIs are problematic: Applications heavily relying on asynchronous callbacks (like K-9 Mail) pose significant challenges for current offloading frameworks, often requiring substantial code rewrites.
Key Results
- AndTweet Case Study: Offloading communication tasks in AndTweet resulted in a 25% reduction in total energy consumption for a single timeline refresh event over Wi-Fi. However, execution time slightly increased.
- 3G Performance: In a 3G network, offloading increased energy consumption due to higher latency and longer radio inactivity timers, despite reduced data transmission.
- Constraint Statistics: Analysis of 16 open-source apps revealed that, on average, only 0.17% of methods are directly migratable without any changes. With minor modifications, up to 47% of methods could potentially be made offloadable.
- Tool Development: The paper presents two functional tools: one for identifying offloading constraints and another for profiling communication energy consumption.
Limitations and Future Work
- Limitations:
- The study primarily focuses on Wi-Fi and 3G networks, and the findings might not generalize to other network types.
- The ThinkAir framework has its own limitations, particularly with asynchronous APIs, which are not universally addressed by other frameworks.
- The constraint identification tool is based on heuristics and might not catch all possible issues.
- The energy profiling tool relies on models that make assumptions about interdependencies, leading to energy estimates presented as ranges.
- Future Work:
- The authors envision a complete toolkit that integrates constraint identification, energy profiling, and a "structure analyzer" to guide larger code refactoring for offloading.
- They plan to integrate CPU usage measurement and estimation into their toolkit, as computation can also be a significant energy factor.
- Further research is needed to address challenges with threading and complex program execution flows.
- Developing efficient automatic synchronization mechanisms for state management between the device and surrogate is crucial.
Summary updated