Research On Resource Scheduling For Mobile Cloud Workflows

As a novel integration technology of the mobile computing and cloud computing,mobile cloud computing(MCC)provides an effective and promising solution for the mobile devices with restricted hardware resources,insufficient processing ability and limited battery life to improve their service quality and alleviate the resource constraint problem.Although MCC based application is becoming more and more popular,it still feces lots of issues and challenges.Particularly,in the mobile cloud workflow systems,it is very difficult to take full advantage of the cloud computing due to the lack of in-depth understanding of the way of energy consumption and effective resource scheduling strategy.It is an urgent problem to optimize the power management and design a practical and efficient resource scheduling strategy for mobile cioud workflows.In terms of the power management and workflow resource scheduling in MCC,a large amount of related works has been proposed at home and abroad.As for the power management,the existing works mainly involve power prediction and reduction.However,the monitor objects of these methods focus on the system performance and little work has been devoted to mining the energy consumption characteristics of the application itself.Computation offloading is another important research hotspot for power management in MCC,which involves the design of real-time system frameworks and offloading algorithm prototypes.However,most of these works only apply to the single applications,not to the workflow systems.As for the workflow scheduling,the existing works mainly involve best-effort based scheduling and QoS-constraints based scheduling strategies,while these resource scheduling strategies are more applicable to the grid computing or cloud computing but may not apply to the mobile cloud workflow.Towards the challenges and problems mentioned above,the basic idea for this thesis research is to optimize the intergration of the mobile cloud workflow performance and the mobile device energy consumption.Concentrated on the key problem,the research content carried on in this thesis can be summarized and listed as follows:(1)Towards the mobile cloud workflow activities,a time-series based prediction approach is proposed for the power management.With the aid of time-series thinking,the approach uses the real-time power sequence formed during the execution process of the activity to predict the power directly.As for the activities that have history power data record,the traditional time-series regression models(e.g.ARMA(p,q)(Auto-Regressive and Moving Average Model)or ES(Exponential Smoothing))are used to predict the power.As for the new activities that have no history power data record,the time-series segmentation models(e.g.Sliding Windows Segmentation or Bottom Up Segmentation)are used to predict the real-time power.(2)Towards the mobile cloud workflows,an energy-efficient UtilityCost optimization model is proposed for systematic analyzing and solving the resource scheduling problem First of all,taking full consideration of the execution characteristics of the mobile cloud workflow and the QoS constraints defined by the mobile users,it formulates the resource scheduling problem systematically,including the application model,system model,execution time model and energy consumption model.Then,introduce a new concept"UtilityCost" for mobile device,to optimization the trade-off between the two goals:minimizing the energy consumption of mobile terminal and the total execution time of workflow for the problem.Finally,transform the resource scheduling problem into an optimization problem with the goal of minimizing "UtilityCost" under QoS constraints.(3)Towards the mobile cloud workflows,an energy-efficient resource scheduling approach named "EERA" is proposed to make the scheduling decision for each activity.Based on the mapping relationship between activities and resources,the optimization scheduling model depends on the resource scheduling solution which can be obtained by the EERA.EERA is a three-phase and meta-heuristic based resource scheduling approach for mobile cloud workflow.Specifically,in phase-one,the workflow execution process has been divided into several partial critical paths(PCPs);in phase-two,an improved discrete particle swarm optimization algorithm(shorted for DPSO)is employed to find the local optimal resource scheduling solution with the minimum UtilityCost on each path;in phase-three,the global optimal resource scheduling solution can be obtained by the integration of all the local optimal solutions.For the above research content,this thesis designs a real-world mobile cloud workflow on our mobile cloud computing environment based on OpenStack and experimental results validate the effectiveness of the proposed approaches mentioned above.