Improving efficiency, scalability and efficacy of adaptive computation offloading in pervasive computing environments

Teshome Abebe, E 2012, Improving efficiency, scalability and efficacy of adaptive computation offloading in pervasive computing environments, Doctor of Philosophy (PhD), Computer Science & IT, RMIT University.

Document type: Thesis
Collection: Theses

Attached Files
Name Description MIMEType Size
Abebe.pdf Thesis Click to show the corresponding preview/stream application/pdf;... 21.58MB
Title Improving efficiency, scalability and efficacy of adaptive computation offloading in pervasive computing environments
Author(s) Teshome Abebe, E
Year 2012
Abstract As computing becomes more mobile and pervasive, there is a growing demand for increasingly rich, and therefore more computationally heavy, applications to run in mobile spaces. However, there exists a disparity between mobile platforms and the desktop environments upon which computationally heavy applications have traditionally run, which is likely to persist as both domains evolve at a competing pace. Consequently, an active research area is Adaptive Computation Offloading or cyber foraging that dynamically distributes application functionality to available peer devices according to resource availability and application behaviour.

Integral to any offloading strategy is an adaptive decision making algorithm that computes the optimal placement of application components to remote devices based on changing environmental context. As this decision is typically computed by constrained devices and may occur frequently in dynamic environments, such algorithms should be both resource efficient and yield efficacious adaptation results. However, existing adaptive offloading approaches incur a number of overheads, which limit their applicability in mobile and pervasive spaces.

This thesis is concerned with improving upon these limitations by specifically focusing on the efficiency, scalability and efficacy aspects of two major sub processes of adaptation: 1) Adaptive Candidate Device Selection and 2) Adaptive Object Topology Computation. To this end, three novel approaches are proposed.

Firstly, a distributed approach to candidate device selection, which reduces the need to communicate collaboration metrics, and allows for the partial distribution of adaptation decision-making, is proposed. The approach is shown to reduce network consumption by over 90% and power consumption by as much as 96%, while maintaining linear memory complexity in contrast to the quadratic complexity of an existing approach. Hence, the approach presents a more efficient and scalable alternative for candidate device selection in mobile and pervasive environments.

Secondly, with regards to the efficacy of adaptive object topology computation, a new type of adaptation granularity that combines the efficacy of fine-grained adaptation with the efficiency of coarse level approaches is proposed. The approach is shown to improve the efficacy of adaptation decisions by reducing network overheads by a minimum of 17% to as much 99%, while maintaining comparable decision making efficiency to coarse level adaptation.

Thirdly, with regards to efficiency and scalability of object topology computation, a novel distributed approach to computing adaptation decisions is proposed, in which each device maintains a distributed local application sub-graph, consisting only of components in its own memory space. The approach is shown to reduce network cost by 100%, collaboration-wide memory cost by between 37% and 50%, battery usage by between 63% and 93%, and adaptation time by between 19% and 98%.

Lastly, since improving the utility of adaptation in mobile and pervasive environments requires the simultaneous improvement of its sub processes, an adaptation engine, which consolidates the individual approaches presented above, is proposed. The consolidated adaptation engine is shown to improve the overall efficiency, scalability and efficacy of adaptation under a varying range of environmental conditions, which simulate dynamic and heterogeneous mobile environments.
Degree Doctor of Philosophy (PhD)
Institution RMIT University
School, Department or Centre Computer Science & IT
Keyword(s) Distributed Computing
Adaptive Software Systems
Adaptive Computation Offloading
Application Adaptation
Application Partitioning
Application Graph
Class Graph
Hybrid Granularity Graph
Object Graph
Graph Partitioning
Mobile Objects
Mobile Code
Version Filter Type
Access Statistics: 283 Abstract Views, 423 File Downloads  -  Detailed Statistics
Created: Mon, 03 Dec 2012, 10:49:50 EST by Brett Fenton
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us