没有合适的资源?快使用搜索试试~ 我知道了~
云计算-移动边缘网络计算卸载调度与资源管理策略优化研究.pdf
1.该资源内容由用户上传,如若侵权请联系客服进行举报
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
2.虚拟产品一经售出概不退款(资源遇到问题,请及时私信上传者)
版权申诉
0 下载量 22 浏览量
2022-07-07
05:17:36
上传
评论 1
收藏 2.93MB PDF 举报
温馨提示
试读
146页
云计算-移动边缘网络计算卸载调度与资源管理策略优化研究.pdf
资源推荐
资源详情
资源评论
摘要
摘 要
通过将计算和存储资源广泛地分布到更接近用户或数据源的网络边缘,移动
边缘计算(Mobile Edge Computing, MEC)支持在无线接入网内完成移动应用的计
算卸载过程,大幅降低 了网络的端到端时延,并有效减轻了核 心网和数据中心的
处理 压力。计算卸载决策,包括用户侧卸载决策(如任务是否卸载、如何卸载以
及何 时卸载)和运营商侧卸载决策(如是否允许用户卸载、分配多少资源进行卸
载), 是MEC能否提升用户体验的关键。由于MEC环境的复杂性,影响卸载决策
的因素 众多,如何设计最优的卸载决策策略,以充分 挖掘MEC在时延、能耗上的
性能增益,是非常具有挑战性的科学问题。
任务调度与资源管理是MEC卸载决策过程中需要考虑的两个重要因 素。一方
面,MEC环境本质上是一个分布式异构并行计算环境,只有对任务进行合理调度,
才能充分发挥该计算环境的性能优势。若考虑网络环境的动态变化,则 还需要对
任务 卸载的时机进行调度决策。另一方面,由于网络边缘的资源有限,必须对这
些资源进行合理分配以充分发挥它们的 效用。在 用户较多的情况下,还需要对是
否允许用户卸载进行决策(即准入决策), 避免资源的过度竞争。在此背景下,本
文分别针对用 户视角下的面向图依赖关系的任务卸载调度和面向复杂任务队列的
动态任务卸载调度,以及运营商视角下的考虑用户移动性的准入决策与资 源分配
三个场景,考虑静态环境与动态环境中不同应用卸载 模型的影响,探索MEC计算
卸载调度与资源管理的最优策略。
首先,从移动用户视角,针对 静态场景下具有有向无环图(Directed Acyclic
Graph, DAG)依赖关系的任务的卸载调度决 策问题,充分考虑 网络边缘计算和通
信资源均受限的情况,本文提出了一种基于深度强化学习(Deep Reinforcement
Learning, DRL)的通用型DAG任务卸载调度算法,分别实现了执行时延最小与
用户效用最大两种目的下的卸载调度决策。具体来说,将用户对DAG任务的卸载
调度决策过程建模为马尔可夫决策过程(Markov Decision Process, MDP); 设计
了基于循环神经网络的序列到序列参数共享神经网络架构,以及相应的DAG嵌
入方法, 用于拟合该MDP的卸载调度策略; 并采用当前最优的近端策略优化
(Proximal Policy Optimization, PPO)来完成对该策略网络的训练。通过在不 同环
境及不同卸载调度目的下与六种基线算法的充分对比,验证了所提算法的有效性
和可靠性。
I
万方数据
摘要
进一步,仍然基于移动用户视角,本文针对高动态场景下的车辆MEC计算卸
载调 度决策问题进行了研究,充分考虑 任务到达、任务属性、无线信道、以及用
户移动等动态因素。由于复杂的动态环境和巨大的状态/解 空间,这个随 机优化问
题很难采用传统的优化方法求解。设计了一种基于DRL的面向复杂任务队列的动
态计 算卸载调度算法,联合求解“何处”与“何时”对任务队列中的每个任务进
行卸载调度,以获得在该 复杂环境下任务执行时延与能耗的最优长期折中。采用
了一系列方法来提高该算法的 训练效率和收敛性能,包括利用PPO保证训练过程
的高效和稳定、将卷积神经网络嵌入到策略网络中以提取任务队列的关键特征、
通过对状态和奖励的精细控制避免训练过程中过多的低效探索等。大量仿真结果
证明,所提算法能够在不同环境和不同用户偏好下,获得远高于传统基线算法的
性能。
最后,从运营商视角,本文研究了车载MEC场景下多个运动中的用户之间的
准入决策和资源分配策略,充分考虑计算与通信资源限制、任务截止 时延要求以
及用户移动性的约束,旨在最大化全局系统效用。将该问题建模为一个非线性混
合整数规划问 题,提出了一种具有多项 式时间复杂度的启发式多用户移动 感知卸
载决策算法。通过参数控制的方式将该全局优化问题转化为有限个局部优化问题,
进而将每个局部优化问题分解为一个凸子问题和一个非线性 整数 规划子问题。针
对计算资源分配的凸子问题,采用了数值方式进行求解;而针对确定 准入对象的
非线性整数规 划子问题,则设计了一种基于偏序的启发式方法来获得其 近似最优
解。最终,全局优化问 题的解可以通过求解有限数量的局部优化问题而得到。通
过与六种基线算法的全面对比,验证了所提算法优异的性能和稳定性。
关键词:移动边缘计算,计算卸载,任务调度,资源分配,深度强化学习
II
万方数据
ABSTRACT
ABSTRACT
By distributing computation and storage resources to the network edge in the vicinity
of users and data sources, mobile edge computing (MEC) supports mobile applications
to complete their computation offloading process within the radio access network. Such
a novel computing architecture significantly reduces the end-to-end network latency and
effectively releases the burden of the core network and data center. Offloading decision-
making, including user-side decision-making (e.g., whether to offload, how to offload,
and when to offload) and operator-side decision-making (e.g., offloading request admis-
sion and resource allocation), is the key to efficiently utilize MEC. Due to the complexity
of MEC environments, the decision-making process is affected by many factors. How to
design the optimal offloading decision strategy to fully exploit the potential of MEC in
terms of latency and energy consumption is a very challenging scientific problem.
Task scheduling and resource management are two crucial factors in MEC offload-
ing decision process. On one hand, the MEC environment is essentially a distributed
heterogeneous parallel computing environment. Only with reasonable task scheduling,
the potential of such a computing environment can be fully exploited. When considering
the dynamics of the wireless network, it is necessary to properly determine the timing
of task scheduling. On the other hand, the limited resources deployed at the network
edge need to be allocated appropriately to maximize their utility. It is also necessary to
perform admission decisions among offloading requests to avoid excessive resource con-
tention. To this end, the dissertation focuses on three offloading decision scenarios (i.e.,
the offloading scheduling of tasks with graph dependency in a static environment from
the perspective of users, the offloading scheduling of tasks in a complex task queue in a
dynamic environment from the perspective of users, and the offloading admission and re-
source allocation considering user mobility from the perspective of operators) to explore
the optimal strategy for offloading scheduling and resource management.
First, the dissertation studies the decision-making problem of offloading scheduling
on tasks with directed acyclic graph (DAG) dependency in a static environment from the
perspective of users, while fully considering the limited computation and communication
resources at the network edge. A general DAG-oriented offloading scheduling algorithm
based on deep reinforcement learning (DRL) is proposed, which can achieve two different
III
万方数据
ABSTRACT
offloading scheduling goals: minimizing execution latency and maximizing user utility.
Specifically, the offloading scheduling decision process on tasks with DAG dependency
is modeled as a Markov decision process (MDP). A recurrent neural network-based se-
quence to sequence parameter-shared deep neural network architecture, as well as the
corresponding DAG embedding method, are proposed to approximate the MDP offload-
ing scheduling policy, which is then trained based on state-of-the-art proximal policy
optimization (PPO). The effectiveness and reliability of the proposed algorithm are ver-
ified through comparison with six baseline algorithms in different environments and for
different offloading scheduling goals.
Furthermore, also from the perspective of users, the offloading scheduling decision-
making problem in a highly dynamic vehicular MEC environment is studied, in which
all the dynamic factors, such as task arrival, task attribute, wireless channel, and user
mobility, are taken into account. This stochastic optimization problem is very difficult
for conventional solutions because of the sophisticated environment dynamics and vast
state space. A DRL-based dynamic offloading scheduling algorithm is designed, which
jointly solves “where” to schedule and “when” to schedule each task in the complex task
queue, to achieve the optimal long-term tradeoff between task latency and energy con-
sumption in such a complicated environment. A series of methods are adopted to improve
the training efficiency and convergence performance of the algorithm, including utilizing
PPO to ensure the efficiency and stability of the training process, embedding a convolu-
tional neural network into the policy network to extract the key features of the complex
task queue, and adjusting the state and reward of DRL to avoid inefficient exploration in
the training process. Extensive simulation experiments demonstrate that the performance
of the proposed algorithm is much higher than that of traditional baseline algorithms in
different environments and user preferences.
Finally, from the perspective of operators, the admission decision and resource al-
location strategy among multiple moving users in a vehicular MEC environment is in-
vestigated, in which the constraints of computation and communication resource limi-
tation, task deadline requirements, and the mobility of users are taken into account, in
order to maximize the overall system utility. The optimization problem is formulated as
a mixed-integer non-linear programming (MINLP), and a heuristic multi-user mobility-
aware offloading decision algorithm with polynomial time complexity is proposed. The
original global optimization problem is converted into a finite number of local optimiza-
IV
万方数据
ABSTRACT
tion problems through parameter control, and each local optimization problem is then
decomposed into a convex subproblem and a non-linear integer programming (NLIP)
subproblem. The convex subproblem is solved with a numerical method to obtain the
optimal resource allocation, and a partial order based heuristic approach is designed for
the NLIP subproblem to determine the approximate optimal offloading decision. Finally,
the solution to the global optimization problem is obtained by solving all the local op-
timization problems. Extensive simulation experiments and comprehensive comparison
with six baseline algorithms demonstrate the excellent performance of the proposed al-
gorithm.
Keywords: Mobile edge computing, computation offloading, task scheduling, resource
allocation, deep reinforcement learning
V
万方数据
剩余145页未读,继续阅读
资源评论
programxh
- 粉丝: 17
- 资源: 1万+
上传资源 快速赚钱
- 我的内容管理 展开
- 我的资源 快来上传第一个资源
- 我的收益 登录查看自己的收益
- 我的积分 登录查看自己的积分
- 我的C币 登录后查看C币余额
- 我的收藏
- 我的下载
- 下载帮助
安全验证
文档复制为VIP权益,开通VIP直接复制
信息提交成功