Collaborative Research: CNS Core: Small: Robust Resource Planning and Orchestration to Satisfy End-to-End SLA Requirements in Mobile Edge Networks

移动边缘计算的出现是为了解决云计算作为基于互联网的服务在支持现代移动应用程序时存在的端到端长延迟、低吞吐量和不可预测性等问题。然而，缺乏服务水平协议（sla）形式的性能保证可能导致关键应用程序的性能下降，使其无法使用或不安全。由于移动环境中的高度动态性，边缘提供商可能会因提供可能被违反的SLA保证而招致巨大的财务风险。这阻碍了边缘提供商在没有首先了解并能够控制相关风险的情况下提供SLA保证。该项目旨在开发工具，帮助边缘提供商通过资源规划和编排，在关键性能指标上提供边缘SLA保证，从而量化并最小化相关风险。该项目将大大提高我们对边缘计算风险因素的认识，并为边缘计算研究开辟新的领域。此外，该项目将启用和增强改变生活的边缘应用，如移动视觉和自动驾驶，促进边缘计算行业的投资和加快发展，为未来的计算劳动力培养高素质人才，并通过课程开发和研究传播扩大对边缘计算的认识和兴趣。本项目将为边缘SLA保障的全面风险建模和优化奠定理论和算法基础。本项目将移动边缘应用的实际性能模型与投资组合管理中已建立的风险管理理论相结合，并通过随机优化、凸优化、抽样技术、近似和基于学习的方法开发有效的风险评估和优化算法。具体而言，本项目的技术贡献如下：1)为移动边缘计算应用开发和验证现实的性能和SLA模型；2)为单用户风险感知边缘资源编排建模和优化三种风险度量（风险概率、风险价值和条件风险价值）；3)为多用户单节点风险感知边缘资源编排建模和优化；4)为多用户多节点风险感知边缘资源配置建模和优化。所有的研究成果都将通过试验台和/或大规模的公共模拟进行评估，并将在pi的网站上公开，以促进结果的复制和未来的研究进展。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Mobile edge computing has emerged to address the long end-to-end latency, low throughput, and unpredictability of cloud computing as an Internet-based service when supporting modern mobile applications. Nevertheless, the lack of performance guarantees in the form of service-level agreements (SLAs) can lead to performance degradation of critical applications, rendering them incompetent or unsafe to use. Due to the high dynamics in the mobile environment, the edge provider can incur substantial financial risks for providing SLA guarantees that could be violated. This discourages edge providers from providing SLA guarantees without first understanding and being able to control the associated risks. This project seeks to develop tools that help the edge provider to quantify and minimize the risks associated with providing edge SLA guarantees on key performance metrics through resource planning and orchestration. This project will significantly advance our knowledge on the risk factors in edge computing and give rise to new frontiers in edge computing research. Also, this project will enable and enhance life-changing edge applications such as mobile vision and autonomous driving, promote investment and expedite development in the edge computing industry, train highly qualified personnel for the future computing workforce, and broaden awareness and interest in edge computing through curriculum development and research dissemination.This project will lay the theoretical and algorithmic foundation of comprehensive risk modeling and optimization for providing edge SLA guarantees. This project combines a realistic performance model of mobile edge applications with the established theory of risk management in portfolio management and develops efficient algorithms for risk assessment and optimization through stochastic optimization, convex optimization, sampling techniques, approximations, and learning-based methods. Specifically, this project makes the following technical contributions: 1) development and validation of a realistic performance and SLA model for mobile edge computing applications, 2) modeling and optimization of three risk measures (risk probability, value-at-risk, and conditional value-at-risk) for single-user risk-aware edge resource orchestration, 3) modeling and optimization for multi-user single-node risk-aware edge resource orchestration, and 4) modeling and optimization for multi-user multi-node risk-aware edge resource provisioning. All research outcomes will be evaluated with testbed and/or large-scale simulations with public traces, and will be made publicly available on the PIs' websites to promote result reproduction and future research advancements along the line.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.