Batch Scheduling for Energy-Efficient Sensing in Smartphones

Sensing in smartphones consumes a significant amount of energy and leads to quick depletion of the battery. Most of the existing solutions to overcome the short battery lifetime caused by periodic sensing are personalized. They tend to learn and predict the user activities. Thus, fewer samples are required to recognize user state and sensing intervals can be extended. However, such methods require a training phase and any change in the user pattern causes a need for a new training phase. Therefore, in addition to personalized learning methods, we also need user-agnostic techniques that guarantee instant energy savings independently of the context to be recognized. In this thesis a user-agnostic method that seeks to provide energy efficiency in sensing is proposed. Our approach is based on the observation that an energy overhead occurs every time the CPU is woken up to perform a sensor sampling task. Hence, our goal is to decrease the number of CPU wake-ups incurred due to periodic sampling by combining multiple sensing actions into one joint activity. Our contribution is a mechanism that batches the execution of periodic tasks. BASS (BAtch Scheduler for Sensing) uses the greatest common divisor of the time intervals defined for the sensor sampling tasks. It also introduces a flexibility factor that implies the time delay tolerance regarding the execution of a task. Moreover, our tool implements a detection method for CPU wake-ups caused by any other application or the user. Based on the above, BASS applies batch scheduling to execute the sensor sampling tasks in batches and result in fewer CPU wake-ups. We evaluated our mechanism using a sensing application for monitoring patients that suffer from Rheumatic Arthritis. We conducted a number of experiments on an HTC Sensation phone, which showed that the efficient exploit of CPU wake-ups cuts down the energy consumption in mobile sensing. The BASS tool achieved an average power reduction of up to 44% and 18% in laboratory and real-world experiments respectively, in our application scenario, without compromising sensing time accuracy.

Subject

smartphone
batching
energy efficiency
scheduler

To reference this document use:

http://resolver.tudelft.nl/uuid:8ee5c6ed-63dd-49dc-b62e-e6c9b5c2b876

Embargo date

2013-08-02

Part of collection

Student theses

Document type

master thesis

Rights

Files

PDF

Kimon_Tsitsikas_Thesis.pdf

3 MB

Close viewer