SHF: SMALL: STITCH: Request-SLO-Aware Orchestration for Large-scale Sensing Services over IoT-Edge-Cloud Hierarchy

随着三层物联网边缘云层次结构在社会和地理规模以及传感、计算和存储能力方面发展成为一个成熟的生态系统，云正在扩展到全球的每一个角落。这为开发一种全新的云服务类别提供了机会，即基于物联网的感知即服务（ISaaS）。有了ISaaS，用户可以用指尖实时“感知”世界的任何部分，甚至整个世界。从资源需求、社会、经济和政治意义等方面来看，预计涉及多达数十亿个边缘和物联网设备的大规模ISaaS服务将成为云服务的主导类别。值得注意的例子（按时间重要性降序排列）是地震检测和警报、儿童/病人/物体跟踪和识别、用于病毒爆发检测的国家/全球发烧/症状映射、环境和公用事业监测以及用于业务分析的人群外包。显然，要启用具有不同响应时间需求的广泛ISaaS服务，技术上称为服务水平目标（SLO），必须有一个通用的编排平台，可以在这些服务之间协调资源分配，并为这些服务提供SLO保证，包括整个物联网边缘云生态系统。该项目旨在开发这样一个平台，称为基于物联网边缘云层次（STITCH）的大规模传感服务的请求-慢速感知资源编排。STITCH采用的方法促进了基础分析，以指导稳健复杂系统的设计，并提供了新的云服务。它将有助于促进学术界和工业界在设计有效的大规模计算系统方面的合作。UTA是一所西班牙裔服务机构，通过现有的课程设置和REU项目，参与这项研究的少数族裔和女性学生以及残疾学生将有助于提高未来美国劳动力的竞争力。STITCH是一个两级分布式资源编排平台，即云到边缘集群级和边缘集群到边缘和物联网级。STITCH的核心是开发一种新的分解技术，该技术可以将给定的请求SLO转换为请求的所有感知任务的分布式任务性能预算。更具体地说，在云中，分解技术将给定ISaaS服务的请求SLO转换为边缘集群级的任务响应时间预算，用于将请求的感知任务分派到不同的边缘集群。反过来，在边缘集群级别，分解技术将每个任务的边缘集群级别任务响应时间预算转换为任务进一步分配到的单个边缘和物联网设备的任务队列截止日期。该解决方案使STITCH能够以完全分布式的方式调度传感任务，从而可以扩展到数十亿个边缘和物联网设备，同时提供请求SLO保证。提出的研究解决了大规模实现ISaaS服务的关键挑战。首先，与现有的主要以资源为中心的ISaaS编排平台不同，STITCH是一个整体的、以用户为中心的解决方案，它为任何ISaaS服务的单个用户请求提供SLO保证。其次，与现有的云中的请求/作业调度直接与边缘和物联网资源分配有关的解决方案不同，在STITCH中，云中的请求调度纯粹由高级用户需求驱动，而任务资源分配则由各个边缘集群处理。这种关注点的分离使得STITCH具有高度可扩展性，并且可以在边缘进行自主控制。最后，所采取的方法是基础的，应用基本原理和数学推理来解决实际重要性的挑战。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。

As the three-tier IoT-Edge-Cloud hierarchy is evolving into a mature ecosystem in terms of its social and geographical scale, and sensing, computing, and storage capabilities, the cloud is expanding its reach to every corner of the globe. This gives rise to the opportunity of developing a whole new category of cloud services, known as IoT-based Sensing-as-a-Service (ISaaS). With ISaaS, a user would be able to ”sense” any part of the world or even the entire world at his/her fingertip in real time. It is expected that large-scale ISaaS services involving up to billions of edge and IoT devices for simultaneous sensing will emerge as a dominant category of cloud services, in terms of resource demand, and social, economic, and political significance. Notable examples, in decreasing order of time criticality, are earthquake detection and alert, child/patient/object tracking and identification, nation-/world-wide fever/symptom mapping for virus breakout detection, environment and utility monitoring, and crowd sourcing for business analytics. Obviously, to enable such w¬ide range of ISaaS services with diverse response time requirements, technically known as service level objectives (SLOs), a common orchestration platform that can coordinate resource allocation among and provide an SLO guarantee for such services, encompassing the entire IoT-Edge-Cloud ecosystem, must be in place. This project aims at developing such a platform, called requeST-SLO-aware resource orchestration for large-scale sensing services over IoT-edge-Cloud Hierarchy (STITCH). The approach taken by STITCH promotes fundamental analysis in guiding the design of robust complex systems and provides enablers of new cloud services. It will help foster collaboration between academia and industry in designing effective large-scale computing systems. The involvement of underrepresented minority and women students, and students with disability at UTA, a Hispanics Serving Institution, in this research through existing course offerings and a REU program will help enhance the competitiveness of the future US workforce.STITCH is a two-level distributed resource orchestration platform, i.e., the cloud-to-edge-cluster level and the edge-cluster-to-edge-and-IoT level. At the core of STITCH is the development of a novel decomposition technique that can translate a given request SLO into distributed task performance budgets for all the sensing tasks of the request. More specifically, in the cloud, the decomposition technique translates the request SLO for a given ISaaS service into edge-cluster-level task response-time budgets for the sensing tasks of the request to be dispatched to different edge clusters. In turn, at the edge cluster level, the decomposition technique translates the edge-cluster-level task response-time budgets for each task into task queuing deadlines at individual edge and IoT devices the task is further dispatched to. This solution makes it possible for STITCH to schedule the sensing tasks for sensing in a fully distributed manner, so that it can scale up to billions of edge and IoT devices, while providing request SLO guarantee. The proposed research addresses key challenges to enable ISaaS services at scale. First, unlike the existing ISaaS orchestration platforms that are mostly resource centric, STITCH is a holistic, user-centric solution that provides SLO guarantee for individual user request of any ISaaS service. Second, unlike the existing solutions where request/job scheduling in the cloud is directly concerned with the edge-and-IoT resource allocation, in STITCH, request scheduling in the cloud is purely driven by the high-level user requirements, leaving the task resource allocation to be handled by the individual edge clusters. This separation of concerns makes the STITCH highly scalable and the autonomous control at the edge possible. Finally, the approach taken is foundational, applying fundamental principles and mathematical reasoning to address challenges of practical importance.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.