Application partitioning algorithms in mobile cloud computing: Taxonomy, review and future directions

Abstract

The execution of mobile application using computational cloud resources is the latest augmentation strategy for resources constraint Smart Mobile Devices (SMDs). Computational offloading requires mobile application to be partitioned during the execution of the application on SMDs. Since an optimal partitioning approach promises optimization of energy savings and performance on SMDs, partitioning of mobile application at runtime is a challenging research perspective. This paper reviews existing Application Partitioning Approaches (APAs) from different domains including Mobile Cloud Computing (MCC). This work is driven by the objective to highlight the issues and challenges associated with the existing APAs in dealing with different context, utilizing the local and cloud resource during partitioning, and augmenting the execution of computation intensive mobile application. We proposes thematic taxonomy of current APAs, reviews current partitioning approaches by using thematic taxonomy, and investigates the implications and critical aspects of current partitioning approaches. The commonalities and deviations in such approaches are analysed based on significant parameters such as context-awareness, granularity level, annotation, and partitioning model. Finally, we put forward the open research issues in application partitioning approaches for MCC that remain to be investigated. I.

A Comparative Study of Code Offloading Techniques and Application Partitioning Methods in Mobile Cloud Computing

“…In the same way several graphs are generated in different applications partitioning algorithms according to the level of granularity used such as module level, method level, object level, class level, task level, thread level, component level, etc. [21] . And partitioning can be static(predefined during development) or dynamic (based on runtime conditions).…”

Context-Aware Mobile Application Task Offloading to the Cloud

“…We can broadly classify partitioning techniques into graph-based, linear programming-based and annotation-based techniques [7].…”

“…Finally, the optimisation solver determines the best partitioning method (among those generated by the analyser) that optimises an objective function (using the cost model generated by the profiler) to be used at runtime. It is worth noting that the graph-partitioning problem is Non-deterministic Polynomial Complete (NPC) and so most efficient techniques require manual annotations to the application by the developers to provide cues to guide the partitioning process [7].…”

“…One approach involves saving data and especially Big Data on the cloud [4,5,6]. Another approach involves offloading execution of a mobile application to the cloud [7]. Researchers have proposed many techniques, frameworks, and development models for this purpose.…”

Towards Distributed Mobile Computing

“…Table I shows further analyzed tools and their corresponding partitioning level. It can be noticed that works originally concentrated at VMs and class level, while in recent years they exploited the method-level partitioning both relying on programmer annotations or on automatic partitioning tools [50]. In what follows, we will provide a taxonomy partition of some noticeable works that perform application partitioning and offloading.…”