SHF: Small: A Distributed Scalable End-to-End Tail Latency SLO Guaranteed Resource Management Framework for Microservices

SHF：Small：分布式可扩展端到端尾部延迟 SLO 保证的微服务资源管理框架

• 批准号: 2226117
• 负责人: Hong Jiang
• 金额: $ 60万
• 依托单位: University of Texas at Arlington
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-10-01 至 2025-09-30
• 项目状态: 未结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2226117&HistoricalAwards=false
• 关键词: SHF; Small; Distributed; Scalable; End

Microsevice architecture typically yields highly scalable, maintainable, testable and loosely coupled systems. For this reason, cloud-computing services are shifting from traditional monolithic architectures to microservice architectures to enable fast development, easy debugging and frequent software updating. A primary design goal of cloud computing is to provide user-centric services that meet diverse end-to-end application query tail-latency Service Level Objective (SLO) requirements, while maximizing system-resource utilization. This proposal aims at developing such an approach, known as Distributed scalable End-to-end Tail lAtency SLO Guaranteed rEsource Management framework for microservices (DETAGEM). DETAGEM explores foundational principles to establish a solid theoretical foundation towards a comprehensive solution for resource allocation with guaranteed per-query tail-latency SLO. The algorithms and tools developed in DETAGEM will be made available to benefit a broad set of researchers, engineers and educators, and thereby advancing research across a range of other areas. Moreover, the obtained results from this project will be integrated into existing undergraduate and graduate curricula on a continuous basis. The involvement of underrepresented students in this proposed research at The University of Texas at Arlington will enhance their competitiveness in the future job market.DETAGEM is a comprehensive distributed scalable solution rooted in fundamental principles. The approach taken by DETAGEM is to (a) derive a mathematical foundation that maps the SLO for Directed Acyclic Graph (DAG) workflows for queries to the task queuing budgets for tasks in the queies at their respective microservices, based on the statistics of unloaded task response time measured at individual microservices at runtime; and (b) develop task-queuing-budget-aware queue-management schemes and resource-auto-scaling schemes based on measured task-queuing budget-violation ratios. The objective is to ensure that all the microservices meet their respective task-queuing budgets, which will then guarantee that the end-to-end query tail-latency SLO for the queries are satisfied. The successful completion of the proposed project is expected to greatly advance the state-of-the-art in terms of methodologies, theories, and practical solutions for the microservice architecture.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.

微服务架构通常会产生高度可扩展、可维护、可测试和松散耦合的系统。因此，云计算服务正在从传统的单体架构转向微服务架构，以实现快速开发、轻松调试和频繁的软件更新。云计算的主要设计目标是提供以用户为中心的服务，满足各种端到端应用程序查询尾延迟服务级别目标（SLO）要求，同时最大限度地提高系统资源利用率。该提案旨在开发一种方法，称为微服务的分布式可扩展端到端尾延迟 SLO 保证资源管理框架 (DETAGEM)。 DETAGEM 探索基本原则，为实现具有保证的每个查询尾部延迟 SLO 的资源分配综合解决方案奠定坚实的理论基础。 DETAGEM 开发的算法和工具将使广大研究人员、工程师和教育工作者受益，从而推动一系列其他领域的研究。此外，该项目获得的成果将持续融入现有的本科生和研究生课程。德克萨斯大学阿灵顿分校的这项拟议研究中代表性不足的学生的参与将增强他们在未来就业市场的竞争力。DETAGEM 是一个植根于基本原则的综合分布式可扩展解决方案。 DETAGEM 采用的方法是 (a) 导出一个数学基础，根据已卸载任务的统计数据，将用于查询的有向无环图 (DAG) 工作流的 SLO 映射到各自微服务的队列中任务的任务排队预算在运行时在各个微服务上测量的响应时间； (b) 根据测得的任务排队预算违规率制定任务排队预算感知队列管理方案和资源自动扩展方案。 目标是确保所有微服务满足各自的任务排队预算，从而保证查询的端到端查询尾延迟 SLO 得到满足。该项目的成功完成预计将极大地推进微服务架构的方法、理论和实际解决方案的最先进水平。该奖项反映了 NSF 的法定使命，经评估认为值得支持利用基金会的智力优势和更广泛的影响审查标准。

项目成果

TailGuard: Tail Latency SLO Guaranteed Task Scheduling for Data-Intensive User-Facing Applications

TailGuard：面向数据密集型用户的应用程序的尾部延迟 SLO 保证任务调度

• 发表时间: 2023-07
• 期刊: the 43rd International Conference on Distributed Computing Systems
• 作者: Wang, Zhijun;Li, Huiyang;Sun, Lin;Rosenkrantz, Todd;Che, Hao;Jiang, Hong
• 通讯作者: Jiang, Hong

User Disengagement-Oriented Target Enforcement for Multi-Tenant Database Systems

多租户数据库系统的面向用户脱离的目标执行

• 发表时间: 2023-11
• 期刊: ACM symposium on cloud computing
• 作者: Li, Ning;Jiang, Hong;Che, Hao;Wang, Zhijun;Nguyen, Minh;Rosenkrantz, Todd
• 通讯作者: Rosenkrantz, Todd