Empirical Evaluation of Dynamic Local Adaptation for Distributed Mobile Applications

Abstract

Distributed mobile applications operate on devices with diverse capabilities, in heterogeneous environments, where parameters such as processor, memory and network utilisation, are constantly changing. In order to maintain efficiency in terms of performance and resource utilisation, such applications should be able to adapt to their environment. Therefore, this paper proposes and empirically evaluates a local adaptation strategy for mobile applications, with 'local' referring to a strategy that operates independently on each node in the distributed application. The strategy is based upon a series of formal adaptation models and a suite of mobile application metrics introduced by the authors in a recent paper. The experiments demonstrate the potential practical application of the local adaptation strategy using a number of distinct scenarios involving runtime changes in processor, memory and network utilisation. In order to maintain application efficiency in response to these changing operating conditions, the system reacts by rearranging the object topology of the application by dynamically moving objects between nodes.

A Hybrid Granularity Graph for Improving Adaptive Application Partitioning Efficacy in Mobile Computing Environments

“…An object distribution or object topology can be computed in one of two different levels of granularity object-level [5,10] and classlevel [4,9]. These levels of granularity describe how the application is abstracted and represented in the decision making process at runtime, which determines the efficiency of the topology computation process and the efficacy of the resulting topology in terms of its ability to meet desired objectives such as reducing inter-node network communication costs.…”

“…In such a scenario, assuming a local adaptation [5] strategy is employed, the constrained node computes an adaptation decision by first dynamically selecting a subset of suitable nodes from the collaboration based on their resource availability and suitability relative to the adapting client and its executing application's requirements. The node then computes an object distribution with the following objectives: − Objective 1: Achieve the requirements of the adapting client (e.g.…”

“…In addition, the increased heterogeneity, dynamism and mobility of this environment provides a context in which traditional applications can benefit from the ability to dynamically adapt at runtime to cope with a wide range of execution scenarios [1]. For instance, an application might reduce its network consumption in a bandwidth scarce environment or spread out its memory usage across multiple devices [2][3][4][5][6][7] in the event that a device runs out of memory.…”

“…This can be done for a number of reasons including application behavior extension, load balancing [5], load mitigation [4,7], 1 Computationally heavy applications have large resource requirements, are traditionally executed in desktop environments, and do not necessarily have inherent distributed execution capabilities.…”

“…Adaptive Object Migration [4][5][7][8][9][10][11] is one such strategy in which a device executes a given application by distributing subsets of its runtime objects to other devices. This can be done for a number of reasons including application behavior extension, load balancing [5], load mitigation [4,7], 1 Computationally heavy applications have large resource requirements, are traditionally executed in desktop environments, and do not necessarily have inherent distributed execution capabilities.…”