A flexible fine-grained adaptive framework for parallel mobile hybrid cloud applications

Mobile devices have become ubiquitous and provide ever richer content and functionality. At the same time, applications are also becoming more complex and require ever increasing amount of computational power and energy. With cloud computing providing unlimited elastic on-demand resources, supporting mobile devices with cloud allows overcoming limitations of mobile devices. This is generally known as Mobile Cloud Computing (MCC) and can be achieved through code offloading that selects computationally or data intensive parts of an application, outsources them to more-resourceful spaces and brings back the final results. While code offloading has been widely studied in the past within the context of distributed systems and grid computing, applying it to current mobile applications requires significant amount of manual changes to existing application codes. An alternative is to outsource the entire application process or the whole virtual machine in which the application is running. This solution assumes running the same code on a more-resourceful system is more efficient, but it is coarse-grained and requires significant amount of data to be transferred. Furthermore, requirements and expectations from mobile applications vary considerably by different users using wide range of mobile devices in various environmental conditions. This diversity in requirements and expectations creates wide range of target offloading goals, ranging from maximizing application performance to minimizing mobile energy consumption. The increased dynamicity and complexity of mobile cloud applications requires open systems that interact with the environment while addressing application-specific constraints, user expectations and hardware limitations. Our goal is to facilitate mobile cloud application development by masking all the complexity of mobile-to-cloud code offloading without requiring application developers to rewrite their code or perform additional manual work. Our focus is on separating the application logic, to be developed by programmers, from the application component configuration and distribution, to be adjusted transparently and dynamically at run-time. Our framework is fine-grained, supporting mobile application configuration and distribution at the granularity of individual components; it is flexible, allowing organizations, application developers, or end-users easily adjust target offloading goal or define policy-driven restrictions on offloading budget, execution quality, or privacy and move around of components without modifying the existing application codes; and it is adaptive, addressing the dynamicity in run-time conditions and end-user contexts. It further supports component distribution in a hybrid cloud environment consisting of multiple public and private cloud spaces. Finally, it provides a new code offloading model that supports fully parallel program execution, where application components located at mobile device and multiple cloud spaces are executed independently but concurrently. The proposed solution can be divided into three main parts: First, a light-weight monitoring system, called Monitor, to capture dynamic environmental parameters and end-user context, profile application resource usage and communications, as well as monitoring availability and performance of cloud resources. Profiling energy consumption per specific application components is primary of importance and requires design and development of a fine-grained automatic energy consumption model, as most mobile devices do not provide any tool for direct measurement of consumed energy and different applications with arbitrary number of components might be running at any time. Second, we design and implement two independent performance-based and energy-based models to enable transparent automatic configuration and distribution of application code and data components that address specific organization, application, and end-user requirements. These models leverage dynamic information from the Monitor on run-time parameters, energy and resource usage of different components, and application characteristics to optimize application performance or mobile energy consumption with respect to a predefined policy. Finally, we design and develop a proof-of-concept framework called IMCM, Illinois Mobile Cloud Management, that embodies the described components to enable fine-grained adaptive application component configuration and distribution, while providing flexibility in terms of adjusting desired target optimization goal or defining additional policy-driven constraints on offloading budget, quality of service per resource, and privacy. Evaluations are carried out using a suite of benchmark applications, including computationally-intensive, I/O-intensive, communication-intensive and combined multi-purpose applications. Compared to sequential execution on a mobile device, these empirical benchmarks using IMCM framework result in speedups or energy-savings factor of over 50 times