SWAVES: Softwarized deployment of services in waves around moving users

SWAVES：围绕移动用户的波浪式服务的软件化部署

• 批准号: 512359107
• 负责人: Professor Dr. Holger Karl
• 金额: --
• 依托单位: Hasso-Plattner-Institut für Digital Engineering
• 依托单位国家: 德国
• 项目类别: Research Grants
• 资助国家: 德国
• 项目状态: 未结题
• 来源: https://gepris.dfg.de/gepris/projekt/512359107?language=en
• 关键词: SWAVES; Softwarized; deployment; services; waves

In many application scenarios, users interact with services deployed in a server infrastructure. To ensure low latency in such interactions, it is often advantageous to deploy these services close to a user; for example, at or near the network's edge in a mobile network – hence the name edge computing. This raises interesting problems in managing such distributed services, e.g., how many instances of a service to run where. These aspects have been extensively researched, but often under simplifying assumptions. For example, it is often assumed that executable code for all services is available everywhere. But this is unrealistic due to limited storage size (especially in Internet-of-Things scenarios), limited data rate to distribute code, and the often small probability that a particular service is required at a particular site. It is hence the goal of this project to research approaches to distribute executable code to edge sites where it is actually needed, balancing distribution overhead and storage space, using, e.g., user mobility predictions. In addition, we will consider different forms of code, ranging from mere source code that is very small but needs building before execution to ready-to-execute virtual machines as well as running code in various standby levels (e.g., cold standby with code being locally available but not running up to active/active replication). To do so, we will investigate both conventional optimization approaches and machine-learning-based ones. We will characterize which schemes can achieve what kinds of tradeoffs under which circumstances (e.g., prediction accuracy). From a practical side, we will provide code characterizations (e.g., build times) and performance profiles for popular services to our SNDZoo, making collected data available to the community at large.

在许多应用场景中，用户与部署在服务器基础设施中的服务进行交互。为了确保此类交互中的低延迟，将这些服务部署在靠近用户的位置通常是有利的；例如，位于移动网络中的网络边缘或附近——因此称为边缘计算。这在管理此类分布式服务时提出了有趣的问题，例如，在哪里运行一个服务的多少个实例。这些方面已被广泛研究，但通常是在简化的假设下进行的。例如，通常假设所有服务的可执行代码随处可用。但这是不现实的，因为存储大小有限（尤其是在物联网场景中）、分发代码的数据速率有限，以及特定站点需要特定服务的概率通常很小。因此，该项目的目标是研究将可执行代码分发到实际需要的边缘站点的方法，平衡分发开销和存储空间，例如使用用户移动性预测。 此外，我们将考虑不同形式的代码，从非常小但需要在执行前构建的纯粹源代码到准备执行的虚拟机以及在各种备用级别运行代码（例如，代码在本地可用但不运行到主动/主动复制的冷备用）。为此，我们将研究传统的优化方法和基于机器学习的方法。我们将描述哪些方案可以在哪些情况下实现哪些类型的权衡（例如，预测准确性）。从实践的角度来看，我们将为我们的 SNDZoo 提供流行服务的代码特征（例如构建时间）和性能配置文件，从而使收集到的数据可供整个社区使用。