Execution Optimization Mechanisms For Edge Computing Empowered Mobile Crowdsensing

Mobile crowdsensing distributes sensing tasks to ubiquitous mobile devices and collects sensing data from various sensors integrated on these devices.It is widely agreed that processing the sensing data at the centralized cloud suffers from low latency and high cost problems.To address such problems,edge computing is considered as a complementary,or even alternative,computing paradigm to cloud.Thanks to the close proximity to the data sources,edge computing not only enables fast sensing data processing,but is also promising in exploring various newly emerged communication technologies like Device-to-Device(D2D)communication.Therefore,there have been increasing research interests in integrating edge computing for mobile crowdsensing.Deploying the crowdsensing services at the network edge can fasten the data collection and processing.Meanwhile,exploring D2 D communications at the network edge,and between the mobile devices,can lower the reliance on cellular communications to lower the communication cost and promote the willingness of mobile crowdsensing participation.However,with the consideration of crowd participation,high mobility,and versatile crowdsensing scenarios,there still exist many issues affecting the crowdsensing efficiency to be addressed: 1)Due to mobility of the participated mobile devices,it is hard to accurately capture the possibilities of D2 D communications,as well as to analyze and optimize the crowdsensing efficiency.2)Solely focusing on either communication optimization or computation optimization cannot fully explore the advantages of resource convergence in edge computing empowered crowdsensing,leading to low efficiency and high cost.3)Without the consideration of sensing data distribution and mobility characteristics,the edge service utility is limited.In D2 D based crowdsensing in edge computing,the mobile devices and edge servers actually form an opportunistic network with multiple sources and multiple destinations.To optimize the crowdsensing efficiency,it is critical to accurately describe the sensing task dissemination and data collection behaviors with the consideration of device mobilities and edge server distributions.To this end,we first conduct stochastic analysis and build Ordinary Differential Equations(ODEs)to describe the opportunistic task dissemination and data collection behaviors.Our proposed framework is able to find out the major factors that affect the crowdsensing efficiency.Then,based on our stochastic analysis framework,we design a time allocation scheme for lifetime limited crowdsensing applications,and analyze how the mobility pattern and edge server distribution affect the crowdsensing efficiency via extensive experiments.Due to the limited computation capabilities on the mobile devices,some sensing data processing tasks must be offloaded to the edge servers.In this case,the offloading decision and the sensing data transmissions shall be jointly optimized to promote the task processing efficiency and reduce the unnecessary energy consumption.Software-Defined Networking(SDN)has been regarded as an effective technology to manage the data flows at the network edges.In this case,besides the constraints from the network bandwidth,the qualityof-sensing requirements,the edge server processing capabilities,the flow table size limitations of the SDN switches are also un-ignorable.As a result,we first derive queuing based models to describe the sensing task processing on both mobile devices and edge servers,and then accordingly form the offloading decision and data flow scheduling for overall energy minimization into an Integer Linear Programming(ILP)problem.To tackle the high computation complexity of solving ILP,a low complexity heuristic algorithm is proposed,achieving 87.4% energy efficiency against to the optimal solution.It is desirable to maximization the utilization of D2 D communications by deploying large number of mobile devices to lower the crowdsensing operational cost.But this may result in high capital cost at the same time.To this end,a service deployment strategy that balances the operational cost and capital cost by exploring the public transportation network and citizen mobility characteristics is proposed.The sensing data collection at the edge services is described as a multi-commodity flow problem,based on which the cost minimization oriented edge service deployment problem is formulated into an ILP form.By analyzing the citizen mobility characteristics to identify the D2 D based data collection opportunities,a edge service deployment strategy is invented.Compared with the optimal solution,the invented strategy only requires 3.02% higher cost,significantly outperforms existing strategies.In summary,we systematically analyze and optimize the main phases in edge computing empowered mobile crowdsensing,with a special emphasis on the exploration of D2 D communications.The proposed strategies can promote the quality-of-sensing,improve the energy efficiency,and reduce the crowdsensing cost.