Overview
The importance of mobile computing is rapidly increasing but the field is held back from its potential by the difficulty of evaluating new ideas. For example, in sharp contrast to 3.3 billion of cellphone users, the largest published study to understand mobile networks from a user's perspective involves only 255 users. The effectiveness of mobile computing systems depend on variable network, application, and user behaviors, many of which are only visible in large-scale systems. Despite the commercial success of devices such as the iPhone, many limitations still exist, e.g., due to resource constraints and mobility challenges, existing applications on such platforms cannot effectively scale to support arbitrary desktop applications. Moreover, the computing mode on smartphones is still quite restricted, with most applications limited to the traditional client-server paradigm. Additional research on networking, applications, and user behavior, as well as the design of mobile devices and systems, is necessary to enable flexible systems using an appropriate mix of centralized and peer-to-peer communication. We believe that research progress in this space is hampered by a lack of visibility into mobile user and system behavior as well as the difficulty of evaluating system ideas in large-scale realistic live use environments.
To enable transformative research on mobile computing and related disciplines, we are working to create a global-scale live laboratory composed of mobile devices, networks, and their users that will support the validation of research ideas. It is our intention that this live laboratory, which we refer to as MobiLab, will be somewhat analogous to the PlanetLab testbed. Unlike PlanetLab, which deploys computation resources to universities and research institutes, MobiLab's initial infrastructure will consist of real mobile users of our previously released smartphone software tools. Our 3GTest software for characterizing cellular network performance and policies and PowerTutor for application-level power and performance estimation have attracted well over 80,000 users across the globe on diverse mobile platforms within just one year. The MobiLab infrastructure should make two key contributions: (1) it provides an online repository of real-time and archived mobile user data, including vital information such as location-specific network conditions, per-application energy usage, and user behavior and (2) through a set of software libraries and tools, it simplifies the tasks of developing, deploying and evaluating research applications. Instead of re-building applications to monitor network, application, mobile system, and user behavior, researchers can gather information online from MobiLab tools via easy-to-use APIs. Our target communities include researchers in several disciplines, such as mobile systems and networks, embedded systems, network security, mobile data privacy, and social networking. We have developed our plan for MobiLab in collaboration with members of these research communities.
Intellectual Merit: The proposed infrastructure enables several classes of research projects that benefit from the following features.
- Mobile system platform support: We will develop tools for real-time characterization of the execution environment including network conditions and energy/power use. These tools serve as middleware for researchers to build prototypes to investigate ideas in mobile system and application design.
- Mobile system resource and computation management: We will provide middleware mechanisms and runtime policies to enable efficient sharing management and sharing of computation, communication, and sensing resources in mobile systems.
- Mobile user data repository: We will build and provide access to a real-time database of (anonymized) mobile user behavior, network activity, and platform information. This will enable trace-driven simulation studies and the design of mobile network systems that use such real-time information for network diagnosis or mobile application adaptation.
Broader Impacts: The proposed infrastructure will be suitable for both undergraduate lab-intensive introductory courses as well as graduate-level courses that expose students to mobile computing, distributed computer system design, and distributed sensing. MobiLab will also have a lasting impact on the research community by providing a central repository for mobile system research data, software artifacts created by the community, and results based on the data. MobiLab will lighten the steep learning curve in mobile computing and engage more researchers in this field.