Joint Collaborative Task Scheduling And Resource Provisioning For Mobile Edge Computing

Mobile edge computing(MEC)networks are emerging as a new computing paradigm,which extend cloud-computing capabilities to the network edge.Cooperations among comput-ing nodes are leveraged to accomplish computationally intensive tasks,thus supporting comput-ing services with extensive interconnections,collaborative interactions,edge intelligence and secure trust.Compared with traditional cloud computing(CC)networks,MEC networks are composed of computing nodes with larger quantity,more diverse types and wider coverage,especially in more complex network environments.While bringing a series of new challenges,MEC networks have gradually become the focus in the industry and academia.Although there have been various MEC application systems,they still lack efficiently collaborative task scheduling strategies,globally optimized resource provisioning methods,intelligent awareness and decision mechanisms.The optimizations for task scheduling and resource provisioning have been well investigated in CC networks,and many efficient methods have been proposed as well.However,it is difficult to directly apply these methods to MEC networks due to separation of the two optimization strategies and lack of considering computing nodes' resource con-straints.Targeting at node cooperation scenarios with dynamic and heterogeneous resources,and multi-hop and social collaboration,we study the optimization theory and methods of joint collaborative task scheduling and resource provisioning in MEC networks.In particular,we design congestion-aware,deadline-aware,trust-aware and social-aware collaborative mecha-nisms,which enable collaborative task scheduling,globally optimized resource provisioning,intelligent awareness and decision for MEC networks.The main results are as follows:(1)Association-based congestion-aware joint collaborative scheduling and provisioning.The limited resources available to individual edge clouds remain to be an important challenge faced by computation offloading in MEC networks.By exploiting edge clouds' dense deploy-ment,we design a collaborative scheduling framework that enables collaborative computing and resource sharing among edge clouds via user association.Considering dynamics of edge resources and workloads,we study the optimality and stability of the location-aware problem for co-provisioning computing and communication resources.We develop a congestion-aware joint collaborative scheduling and provisioning mechanism,which works online without requiring information on future system dynamics.Here congestion awareness manifests itself in long-term network stability(temporal)and global load balancing(spatial).By exploring heterogeneity in users' offloading quality sensitivity,we propose a contract-based pricing mechanism to en-courage users to select association strategies that satisfy congestion constraints.Accordingly,a tradeoff between user local preference and network global optimization is achieved.Extensive simulation results show that our proposed mechanism can improve system profit,guarantee user incentive and maintain network stability.(2)Peer offloading-based multi-service joint collaborative scheduling and provisioning.Peer offloading allows edge clouds within specific geographic regions to form a shared resource pool,realizing flexible resource supply and collaborative task processing.Existing scheduling mechanisms mostly assume that edge clouds can process whatever types of task requests from users,without considering service availability.Especially under limited storage resources,edge clouds can only provision a subset of services at the same time.Taking resource heterogeneity into account,we design a joint collaborative provisioning and scheduling framework(JCPS)that realizes cost reduction by exploiting spatial-temporal diversities in workload and resource cost among edge clouds.We first formulate and solve the problem of spatially strategic optimization without deadline awareness,which is proved NP-hard.With user deadline tolerance,we further study the spatial-temporal optimality of priority-based joint collaborative provisioning and scheduling strategies.We consider both the scenario where edge clouds act cooperatively and the scenario where edge clouds select strategies to solely pursue their own interests.In particular,two deadline-aware joint collaborative mechanisms,D-JCPS and AD-JCPS,are proposed respectively.Extensive simulation results show that they can guarantee near-optimal system performance,and reveal the importance of jointly optimizing collaborative scheduling and provisioning.(3)Multi-hop offloading-based trusted joint collaborative scheduling and provisioning.Multi-hop offloading would be more promising in exploiting collaborative computing capabil-ities,in that it breaks region limitation and dependency on remote cloud.Considering trust risks in MEC collaboration committed by self-interested edge owners,we establish a social trust propagation model,and study the system cost minimization problem under co-provisioning of computing,communication and trust resources.As expected,latency versus trust risk tradeoff is achieved.However,the optimal offloading control can not materialize without complete latency knowledge.Existing scheduling mechanisms mostly focus on the case of deterministic completion latency.But in practical distributed computing systems,task completion latency usually presents high variability mainly due to different workload levels or resource contentions.Therefore,we propose an online learning-aided cooperative offloading mechanism.To adapt to latency variability,we harness the delayed online learning technique to facilitate completion latency prediction,which is used for the basis of offloading decision making.Both theoretical analysis and trace-driven simulations verify that our proposed mechanism guarantees near-optimal system performance and robustness to prediction error.Results also reveal that strong trust propagation can reduce system cost significantly.(4)Device cooperation-based social D2D joint collaborative scheduling and provisioning.Existing scheduling mechanisms mainly focus on resource sharing among edge clouds and remote cloud,without considering the increasing computing power of devices.We study the social D2D collaborative scheduling framework that enables task offloading and resource sharing among devices via D2D direct communications.In particular,the social relationship between users provides effective cooperation incentives.Taking social and strategic collaboration into account,we propose a novel market-based multi-resource allocation mechanism.Specifically,we first introduce a concept called socially equivalent resources,and design multi-resource allocation policy based on DRF fairness.Analysis shows that it satisfies many ideal properties,including sharing incentive,envy-freeness,strategy-proofness and pareto efficiency.Inspired by traditional VCG mechanism,we propose an efficient task offloading policy that achieves truthfulness,individual rationality and robustness.Extensive simulation results show that our proposed mechanism can greatly encourage devices with strong social relationship,high reputation and sufficient resources to share their idle resources.It can further guarantee social fairness-efficiency tradeoff and high quality of task completion.