Mobility-aware computation offloading in edge computing using prediction
Mobility-aware computation offloading in edge computing using prediction
复制标题
DOI：
10.1109/icfec50348.2020.00015
复制
发表时间：
2020-05
期刊：
影响因子：
--
通讯作者：
E. Maleki;Lena Mashayekhy
中科院分区：
文献类型：
--
作者：
E. Maleki;Lena Mashayekhy
边缘计算的一个关键用例是计算卸载，它通过节省能源消耗和减少应用程序的延迟来增强资源受限的移动设备的功能。边缘计算资源（称为“云”）是用户附近资源丰富的计算基础设施，旨在减轻移动设备的过载并提供低延迟服务。将计算卸载到小云的主要挑战是如何有效地将移动应用程序分配给小云，以便分配捕获移动设备固有的移动性，并在应用程序运行时将延迟降至最低。我们通过提出一种新颖的卸载方法来解决这个问题，该方法考虑移动应用程序的动态，包括移动性和不断变化的规格，并将应用程序完全分配给云，同时最大限度地减少其周转时间（延迟和执行时间）。我们首先将问题表述为整数规划模型，以最小化移动应用程序的周转时间。这个问题是一个NP难问题。为了解决这个棘手问题，我们设计了一种称为 OAMC 的计算卸载算法，利用未来的移动应用程序规范，根据我们的预测模型获得智能移动感知卸载决策。我们进行了多次实验来评估我们提出的方法的性能。结果表明，OAMC 在合理的运行时间内实现了接近最佳的周转时间。
A key use case of edge computing is computation offloading that augments the capabilities of resource-constrained mobile devices by conserving their energy consumption and reducing latency of their applications. Edge computing resources, called cloudlets, are resource-rich computing infrastructures nearby users that aim at mitigating the overload of mobile devices and providing low-latency services. A main challenge in computation offloading to cloudlets is how to assign mobile applications to cloudlets efficiently such that the assignment captures the mobility inherent of mobile devices and leads to minimum latency during runtime of the applications. We address this problem by proposing a novel offloading approach that considers dynamics of mobile applications including mobility and changing specifications, and fully assigns applications to cloudlets, while minimizing their turnaround time (latency and execution time). We first formulate the problem as an integer programming model to minimize the turnaround time of mobile applications. This problem is an NP-hard problem. To tackle the intractability, we design a computation offloading algorithm, called OAMC, utilizing future specifications of mobile applications to obtain smart mobility-aware offloading decisions based on our prediction models. We conduct several experiments to evaluate the performance of our proposed approach. The results reveal that OAMC leads to near-optimal turnaround time in a reasonable running time.