TL;DR
This paper tackles the challenge of efficiently managing storage on mobile phones for photo crowdsensing tasks. They propose a novel utility-based strategy (SMPC) that intelligently decides which photos to send or delete to maximize either the photo delivery ratio or minimize average delay, outperforming existing methods.
What Problem Does This Paper Address?
Mobile crowdsensing leverages the sensors on our smartphones to collect data for various tasks. Photo crowdsensing, a specific type, uses phone cameras to capture images for applications like disaster monitoring or mapping. A key bottleneck is the limited storage on mobile devices. Users prefer uploading photos via Wi-Fi (Points of Interest - PoIs) over costly cellular networks. This leads to a scenario where users exchange photos opportunistically to get them to PoIs faster. However, when storage is full, or during limited contact times, users must decide which photos to prioritize for sending or which to delete. Existing methods often lack the intelligence to make these decisions optimally, leading to lower delivery rates or longer delays.
Methods
The core of this work is the Storage Management strategy in mobile phones for Photo Crowdsensing (SMPC). This framework comprises three modules:
- Parameter Collection Module: Gathers essential network parameters like the number of photo copies, the number of devices that have carried a photo, and the number of users who have visited a PoI. This information is exchanged between users and PoIs.
- Utility Calculation Module: This is the heart of the strategy. It calculates a "utility" value for each photo. This utility quantifies how sending or deleting a photo impacts the overall performance goals (delivery ratio or average delay). This is achieved by modeling the performance metrics using differential equations and deriving the photo's utility from their derivatives.
- Decision Module: Based on the calculated utility, this module makes the crucial sending and deleting decisions. Photos with higher utility are prioritized for sending when a connection is available. If storage is full, photos with lower utility are candidates for deletion.
The paper proposes two specific instantiations of SMPC:
- SMPC-R: Prioritizes maximizing the photo delivery ratio.
- SMPC-D: Prioritizes minimizing the average delay.
Datasets
The paper evaluates its proposed strategies using:
- Random-waypoint mobility model: A standard simulation model for user movement.
- Three real-world traces:
- Roma/taxi: GPS data from taxis in Rome, Italy.
- Epfl: GPS data from taxis in San Francisco, CA, USA.
- Geolife: GPS data from users in various locations.
These traces are pre-processed to remove anomalous trajectories.
Key Insights and Ideas
- Utility as a Decision Driver: The paper demonstrates that a well-defined "utility" metric, which quantifies the impact of a photo on system performance, is crucial for making intelligent storage management decisions.
- Decoupled Optimization: SMPC-R and SMPC-D highlight that optimizing for delivery ratio and minimizing delay are distinct goals and require different utility calculations. One cannot simultaneously achieve the absolute best for both.
- Congestion Awareness: The study suggests that in high-congestion scenarios (low storage space), complex strategies like SMPC are necessary, outperforming simpler "delete oldest/newest" policies. In low-congestion scenarios, simpler policies might suffice.
- Parameter Exchange is Key: The effectiveness of SMPC relies on users and PoIs exchanging critical network state information, enabling a more informed decision-making process.
Key Results
The simulations show that:
- SMPC-R consistently achieved the highest delivery ratio across all tested scenarios and datasets.
- SMPC-D consistently achieved the lowest average delay across all tested scenarios and datasets.
- Compared to other baseline strategies (DL-Delete Last, DF-Delete Front, DO-Delete Oldest, DY-Delete Youngest), SMPC-R and SMPC-D significantly outperformed them in their respective optimization goals.
- Delivery ratios generally increase with larger storage space, longer simulation times, and longer photo TTLs.
- Average delay generally increases with larger storage space, longer simulation times, and longer photo TTLs.
Limitations and Future Work
- Impracticality of Centralized Control: The authors acknowledge that relying on a central controller to gather all uncertain variables is impractical for mobile crowdsensing. Their proposed solution relies on distributed parameter collection among users and PoIs.
- Incentive Mechanisms Not Considered: The paper explicitly states that incentive mechanisms for users to participate in crowdsensing are not considered in this work, focusing solely on storage management.
- Uniform Photo Size and Storage: The model assumes all photos are the same size and mobile phones have uniform storage space. Real-world scenarios might involve variable sizes and heterogeneous devices.
- Future Work: The paper suggests exploring how to approximate SMPC's utility in low-congestion regimes with simpler policies that don't require extensive parameter collection and exchange. Further research could also incorporate incentive mechanisms into the storage management strategy.
Summary updated