MEC系统中基于多智能体深度增强学习的协作式任务卸载和资源分配

移动边缘计算(MEC)通过将计算任务卸载到MEC服务器上,在缓解智能移动设备计算负载的同时,可以降低服务时延。然而目前在MEC系统中,关于任务卸载和资源分配仍然存在以下问题：1)边缘节点间缺乏协作;2)计算任务到达与实际环境中动态变化的特征不匹配;3)协作式任务卸载和资源分配动态联合优化问题。为解决上述问题,文章在协作式MEC架构的基础上,提出了一种基于多智能体的深度确定性策略梯度算法(MADDPG)的任务卸载和资源分配算法,最小化系统中所有用户的长期平均成本。仿真结果表明,该算法可以有效降低系统的时延及能耗。     

社会属性感知的边缘计算任务调度策略

边缘计算服务器的负载不均衡将严重影响服务能力,该文提出一种适用于边缘计算场景的任务调度策略(RQ-AIP)。首先,根据服务器的负载分布情况衡量整个网络的负载均衡度,结合强化学习方法为任务匹配合适的边缘服务器,以满足传感器节点任务的资源差异化需求;进而,构造任务时延和终端发射功率的映射关系来满足物理域的约束,结合终端用户社会属性,为任务不断地选择合适的中继终端,通过终端辅助调度的方式实现网络的负载均衡。仿真结果表明,所提出的策略与其他负载均衡策略相比能有效地缓解边缘服务器之间的负载和核心网的流量,降低任务处理时延。     

Resource scheduling in mobile edge computing using improved ant colony algorithm for space information network

With the development of space information network (SIN), new network applications are emerging. Satellites are not only used for storage and transmission but also gradually used for calculation and analysis, so the demand for resources is increasing. But satellite resources are still limited. Mobile edge computing (MEC) is considered an effective technique to reduce the pressure on satellite resources. To solve the problem of task execution delay caused by limited satellite resources, we designed Space Mobile Edge Computing Network (SMECN) architecture. According to this architecture, we propose a resource scheduling method. First, we decompose the user tasks in SMECN, so that the tasks can be assigned to different servers. An improved ant colony resource scheduling algorithm for SMECN is proposed. The heuristic factors and pheromones of the ant colony algorithm are improved through time and resource constraints, and the roulette algorithm is applied to route selection to avoid falling into the local optimum. We propose a dynamic scheduling algorithm to improve the contract network protocol to cope with the dynamic changes of the SIN and dynamically adjust the task execution to improve the service capability of the SIN. The simulation results show that when the number of tasks reaches 200, the algorithm proposed in this paper takes 17.52% less execution time than the Min-Min algorithm, uses 9.58% less resources than the PSO algorithm, and achieves a resource allocation rate of 91.65%. Finally, introducing dynamic scheduling algorithms can effectively reduce task execution time and improve task availability.     

Research on adaptive dual task offloading decision algorithm for parking space recommendation service

In order to improve the efficiency of tasks processing and reduce the energy consumption of new energy vehicle (NEV), an adaptive dual task offloading decision-making scheme for Internet of vehicles is proposed based on information-assisted service of road side units (RSUs) and task offloading theory. Taking the roadside parking space recommendation service as the specific application Scenario, the task offloading model is built and a hierarchical self-organizing network model is constructed, which utilizes the computing power sharing among nodes, RSUs and mobile edge computing (MEC) servers. The task scheduling is performed through the adaptive task offloading decision algorithm, which helps to realize the available parking space recommendation service which is energy-saving and environmental-friendly. Compared with these traditional task offloading decisions, the proposed scheme takes less time and less energy in the whole process of tasks. Simulation results testified the effectiveness of the proposed scheme.     

移动边缘计算卸载技术综述

移动边缘计算(Mobile Edge Computing,MEC)将云服务器的计算资源扩展到更靠近用户一侧的网络边缘,使得用户可以将任务卸载到边缘服务器,从而克服原先云计算中将任务卸载到云服务器所带来的高时延问题。首先介绍了移动边缘计算的基本概念、基本框架和应用场景,然后围绕卸载决策、联合资源分配的卸载决策分别从单MEC服务器和多MEC服务器两种场景总结了任务卸载技术的研究现状,最后结合当前MEC卸载技术中存在的不足展望了未来MEC卸载技术的研究。     

一种面向随机计算卸载的两层无人机能耗优化方法

当物联网设备(Internet of Things Device,IoTD)面临随机到达且复杂度高的计算任务时,因自身计算资源和能力所限,无法进行实时高效的处理。为了应对此类问题,设计了一种两层无人机辅助的移动边缘计算(Mobile Edge Computing,MEC)模型。在该模型中,考虑到IoTD处理随机计算任务时的局限性,引入多架配备MEC服务器的下层无人机和单架上层无人机进行协同处理。为了实现系统能耗最优化,提出了一种资源优化和多无人机位置部署方案,根据计算任务到达的随机性,应用李雅普诺夫优化方法将能耗最小化问题转化为一个确定性问题,应用差分进化(Differential Evolution,DE)算法进行多次变异、交叉和选择取得无人机的优化部署方案;采用深度确定性策略梯度(Depth Deterministic policy Gradient,DDPG)算法对带宽分配、计算资源分配、传输功率分配和任务卸载分配进行联合优化。实验结果表明,该算法相较于对比算法系统能耗降低35%,充分验证了其可行性和有效性。     

Deep reinforcement learning-based joint task offloading and bandwidth allocation for multi-user mobile edge computing

The rapid growth of mobile internet services has yielded a variety of computation-intensive applications such as virtual/augmented reality. Mobile Edge Computing (MEC), which enables mobile terminals to offload computation tasks to servers located at the edge of the cellular networks, has been considered as an efficient approach to relieve the heavy computational burdens and realize an efficient computation offloading. Driven by the consequent requirement for proper resource allocations for computation offloading via MEC, in this paper, we propose a Deep-Q Network (DQN) based task offloading and resource allocation algorithm for the MEC. Specifically, we consider a MEC system in which every mobile terminal has multiple tasks offloaded to the edge server and design a joint task offloading decision and bandwidth allocation optimization to minimize the overall offloading cost in terms of energy cost, computation cost, and delay cost. Although the proposed optimization problem is a mixed integer nonlinear programming in nature, we exploit an emerging DQN technique to solve it. Extensive numerical results show that our proposed DQN-based approach can achieve the near-optimal performance.     

Computing aware scheduling in mobile edge computing system

Mobile edge computing (MEC) is an emerging technology recognized as an effective solution to guarantee the Quality of Service for computation-intensive and latency-critical traffics. In MEC system, the mobile computing, network control and storage functions are deployed by the servers at the network edges (e.g., base station and access points). One of the key issue in designing the MEC system is how to allocate finite computational resources to multi-users. In contrast with previous works, in this paper we solve this issue by combining the real-time traffic classification and CPU scheduling. Specifically, a support vector machine based multi-class classifier is adopted, the parameter tunning and cross-validation are designed in the first place. Since the traffic of same class has similar delay budget and characteristics (e.g. inter-arrival time, packet length), the CPU scheduler could efficiently scheduling the traffic based on the classification results. In the second place, with the consideration of both traffic delay budget and signal baseband processing cost, a preemptive earliest deadline first (EDF) algorithm is deployed for the CPU scheduling. Furthermore, an admission control algorithm that could get rid off the domino effect of the EDF is also given. The simulation results show that, by our proposed scheduling algorithm, the classification accuracy for specific traffic class could be over 82 percent, meanwhile the throughput is much higher than the existing scheduling algorithms.

     

OKRA: optimal task and resource allocation for energy minimization in mobile edge computing systems

To cope with the computational and energy constraints of mobile devices, Mobile Edge Computing (MEC) has recently emerged as a new paradigm that provides IT and cloud-computing services at mobile network edge in close proximity to mobile devices. This paper investigates the energy consumption problem for mobile devices in a multi-user MEC system with different types of computation tasks, random task arrivals, and unpredictable channel conditions. By jointly considering computation task scheduling, CPU frequency scaling, transmit power allocation and subcarrier bandwidth assignment, we formulate it as a stochastic optimization problem aiming at minimizing the power consumption of mobile devices and to maintain the long-term stability of task queues. By leveraging the Lyapunov optimization technique, we propose an online control algorithm (OKRA) to solve the formulation. We prove that this algorithm is able to provide deterministic worst-case latency guarantee for latency-sensitive computation tasks, and balance a desirable tradeoff between power consumption and system stability by appropriately tuning the control parameter. Extensive simulations are carried out to verify the theoretical analysis, and illustrate the impacts of critical parameters to algorithm performance.

     

分层异构信号处理平台调度方法研究

异构多处理器的高效性和可靠性能够满足日趋复杂的信号处理任务需求,因此分层异构系统已成为信号处理平台的发展趋势.为提高平台强实时性并解决高吞吐量的问题,文中对分层异构信号处理平台的软硬件模块及架构进行了研究,并采用有向无环图对组件任务及硬件资源进行建模.将已提出的调度算法按照任务类型、调度目标、调度过程和研究方法进行分类...     

改进的Min-Min网格任务调度算法

在网格环境中,如何对任务进行高效调度是当前研究的热点问题。目前Min—Min调度算法是一个简单、快速、有效的算法。但它很难满足网格任务对服务质量的要求。在独立型的任务调度模型的基础上,提出了一种基于权值的改进Min—Min调度算法。改进后的算法通过量化网格任务的优先级和等待时间,解决了原有算法存在的高质量任务和大任务等待时间过长的问题。仿真实验结果表明,改进后的算法满足了网格任务对优先级和等待时间的服务质量要求．是一种网格环境下有效的任务调度算法。     

军用网格环境下基于优先权的Min-Min任务调度算法

军用网格环境下的资源调度与一般网格环境下的资源调度相比较，一个明显的特点就是必须考虑一些特别任务的优先级。在给出网格独立任务调度模型基础上，提出了一种基于优先权的Min—Min资源调度算法，该算法首先调度优先级高的任务，其余任务则采用Min—Min算法调度。经过分析，该算法的时间复杂度是O（n^2m），与Min—Min相比，该算法的Makespan可能略大，但可以满足军用网格环境下特殊任务优先执行的需求。     

车辆网络多平台卸载智能资源分配算法

为了降低计算任务的时延和系统的成本,移动边缘计算(MEC)被用于车辆网络,以进一步改善车辆服务。该文在考虑计算资源的情况下对车辆网络时延问题进行研究,提出一种多平台卸载智能资源分配算法,对计算资源进行分配,以提高下一代车辆网络的性能。该算法首先使用K临近(KNN)算法对计算任务的卸载平台(云计算、移动边缘计算、本地计算)进行选择,然后在考虑非本地计算资源分配和系统复杂性的情况下,使用强化学习方法,以有效解决使用移动边缘计算的车辆网络中的资源分配问题。仿真结果表明,与任务全部卸载到本地或MEC服务器等基准算法相比,提出的多平台卸载智能资源分配算法实现了时延成本的显著降低,平均可节省系统总成本达80%。     

Task scheduling algorithm for system-wide information management based on multiple QoS constraints

An ant colony optimization task scheduling algorithm based on multiple quality of service constraint (QoS-ACO) for SWIM was proposed.Focusing on the multiple quality of service (QoS) requirements for task requests completed in system-wide information management (SWIM),considering the task execution time,security and reliability factors,a new evaluate user satisfaction utility function and system task scheduling model were constructed.Using the QoS total utility evaluation function of SWIM service scheduling to update the pheromone of the ant colony algorithm.The simulation results show that under the same conditions,the QoS-ACO algorithm is better than the traditional Min-Min algorithm and particle swarm optimization (PSO) algorithm in terms of task completion time,security,reliability and quality of service total utility evaluation value,and it can ensure that the user's task scheduling quality of service requirements are met,and can better complete the scheduling tasks of the SWIM.     

Tasking planning based on task splitting and merging in relay satellite network

To handle the low planning efficiency of the tasks with too long or too short service time,a task planning scheme was proposed based on task splitting and merging for relay satellite systems.First,a task splitting and merging was developed to transfer the task requirements of user to task units which could be planned with high efficiency.Secondly,based on the parallel machine scheduling model,the optimization problem of the task unit planning to maximize the number of completed tasks in the network was built.Further,a heuristic polynomial time scheduling algorithm was proposed.Simulation results show that compared to the traditional scheme,the task planning scheme perform better in terms of completed task number,resource utilization and fairness.     

Integrated Scheduling of Grid Applications in WDM Optical Light-Trail Networks

A novel task graph model, flexible task model (FTM), is proposed for modeling the grid computing tasks and the relationships among the tasks. In this model, a task may generate output before the task completes whereas previous work assumes that no output is available until the task is completed. In addition, a task in FTM can start to execute when it has collected a minimum amount of required input from its predecessors. FTM is more general and flexible than the conventional task graph model considered in previous work. Based on FTM, we investigate the problem of scheduling grid applications that integrates the resource allocation for task execution and service provisioning for subwavelength data communication between the tasks. Data communication between grid tasks under the FTM model is better supported using light-trails in wavelength division multiplexing (WDM) networks, than lightpaths. The objective is to minimize the total amount of time for task completion or makespan. Simulation results show that our proposed scheduling algorithm under FTM significantly reduces the total task completion time compared with that under the conventional task graph model. Moreover, the communication service provisioning using light-trails is very resource efficient.      

Joint computation offloading and task caching for multi-user and multi-task MEC systems: reinforcement learning-based algorithms

Computation offloading at mobile edge computing (MEC) servers can mitigate the resource limitation and reduce the communication latency for mobile devices. Thereby, in this study, we proposed an offloading model for a multi-user MEC system with multi-task. In addition, a new caching concept is introduced for the computation tasks, where the application program and related code for the completed tasks are cached at the edge server. Furthermore, an efficient model of task offloading and caching integration is formulated as a nonlinear problem whose goal is to reduce the total overhead of time and energy. However, solving these types of problems is computationally prohibitive, especially for large-scale of mobile users. Thus, an equivalent form of reinforcement learning is created where the state spaces are defined based on all possible solutions and the actions are defined on the basis of movement between the different states. Afterwards, two effective Q-learning and Deep-Q-Network-based algorithms are proposed to derive the near-optimal solution for this problem. Finally, experimental evaluations verify that our proposed model can substantially minimize the mobile devices’ overhead by deploying computation offloading and task caching strategy reasonably.

     

Negotiated disk admission control in video streaming

A novel soft-quality of service framework for continuous media servers, which provides a dynamic and adaptive disk admission control and scheduling algorithm, is presented. Using this framework, the number of simultaneously running clients for video servers could be increased, and better resource utilisation under heavy communication traffic requirements could be ensured     

云环境下服务信任感知的可信动态级调度方法

以社会学中的人际关系信任模型为基础,提出了一种基于服务消费者的服务满意度评价、推荐者的服务推荐和第三方服务性能反馈的可信度量模型。将用户对服务资源的信任需求和服务资源的可信度并入DLS算法得到可信动态级调度算法CTDLS,从而在计算调度级别时考虑服务资源的可信程度。模拟实验表明,该算法能有效满足任务在信任方面的服务质量需求,对提高任务调度的成功率具有实际意义。     

Deep Q-network-based auto scaling for service in a multi-access edge computing environment

In 5G networks, it is necessary to provide services while meeting various service requirements, such as high data rates and low latency, in response to dynamic network conditions. Multi-access edge computing (MEC) is a promising concept to meet these requirements. The MEC environment enables service providers to deploy their low latency services that are composed of multiple components. However, operating a service manually and attempting to satisfy the quality of service (QoS) requirements is difficult because many factors need to be considered in an MEC scenario. In this paper, we propose an auto-scaling method using deep Q-networks (DQN), which is a reinforcement learning algorithm, to resize the number of instances assigned to service. In our evaluation, compared to other baseline methods, the proposed approach maintains the appropriate number of instances effectively in response to dynamic traffic change while satisfying QoS and minimizing the cost of operating the service in the MEC environment. The proposed method was implemented as a module running in OpenStack and published as open-source software.