CNS Core: Small: Moving Machine Learning into the Next-Generation Cloud Flexibly, Agilely and Efficiently

CNS核心：小：灵活、敏捷、高效地将机器学习迁移到下一代云

• 批准号: 2008265
• 负责人: Dong Dai
• 金额: $ 47.18万
• 依托单位: University of North Carolina at Charlotte
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-07-15 至 2024-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2008265&HistoricalAwards=false
• 关键词: CNS; Core; Small; Moving; Machine

Machine Learning (ML) is being used in new ways to develop intelligent software. Meanwhile, serverless computing is redefining how to use cloud computing platforms. With serverless computing, ML specialists only need to define a set of stateless functions that have access to a common data store. Current ML software systems are generally specialized for first-generation cloud computing systems that do not have the flexibility required for serverless infrastructures. This project aims to take advantage of serverless computing to deploy machine learning software. This simplifies the deployments, avoids the need for infrastructure maintenance, and includes built-in scalability and cost-control. Due to the ease of management and ability to rapidly scale, serverless computation has become the trend to build next-generation ML services and applications. This project proposes a unified serverless computing framework that aims to be flexible, agile and efficient for moving ML into the next-generation cloud to achieve better simplicity, manageability and productivity. In particular, to bridge the semantic gap between the serverful ML model and the serverless cloud platform, this project identifies three major tasks: fine-grained computation management, an efficient communication strategy and a cost-effective service model. This project aims for widespread serverless computing by removing server and operation system level details and simplifying the process of building and managing ML applications. It is a continuum along which developers and operations teams become more accustomed to increased automation and abstraction, and more comfortable breaking ML applications into simple, easy-to- manage microservices, application interfaces, and functions. The result is that developers are free to target the right tools for the right tasks and to build ML applications easily that span any number of different serverless services. This project emphasizes open-source software development. This will enhance the access to stream data analytic frameworks and broaden the project’s impact. Furthermore, the models and workloads/traces from this project may enable further research by others. The project repository[https://abclab-uncc.github.io/website/grants.html] (data, code, results, emulators, and simulators) will be maintained in the next 5 years.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.

机器学习（ML）正在以新的方式用于开发智能软件。 同时，无服务器计算正在重新定义如何使用云计算平台。使用无服务器计算，ML专家只需要定义一组可以访问公共数据存储的无状态函数。当前的ML软件系统通常专用于第一代云计算系统，这些系统不具备无服务器基础设施所需的灵活性。该项目旨在利用无服务器计算来部署机器学习软件。这简化了部署，避免了基础设施维护的需要，并包括内置的可扩展性和成本控制。由于易于管理和快速扩展的能力，无服务器计算已成为构建下一代ML服务和应用程序的趋势。该项目提出了一个统一的无服务器计算框架，旨在灵活，敏捷和高效地将ML迁移到下一代云中，以实现更好的简单性，可扩展性和生产力。特别是，为了弥合服务器ML模型和无服务器云平台之间的语义差距，该项目确定了三个主要任务：细粒度计算管理，高效的通信策略和具有成本效益的服务模型。该项目旨在通过删除服务器和操作系统级别的细节并简化构建和管理ML应用程序的过程来实现广泛的无服务器计算。这是一个沿着的连续体，开发人员和运营团队越来越习惯于增加自动化和抽象，并且更舒适地将ML应用程序分解为简单，易于管理的微服务，应用程序接口和功能。其结果是，开发人员可以自由地将正确的工具用于正确的任务，并轻松构建跨越任何数量的不同无服务器服务的ML应用程序。该项目强调开源软件开发。这将加强对流数据分析框架的访问，并扩大项目的影响。 此外，该项目的模型和工作负载/跟踪可能使其他人能够进一步研究。项目存储库[https：//abclab-uncc.github.io/website/grants.html]（数据、代码、结果、仿真器和模拟器）将在未来5年内维护。该奖项反映了NSF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。

项目成果

Drill: Log-based Anomaly Detection for Large-scale Storage Systems Using Source Code Analysis

• DOI: 10.1109/ipdps54959.2023.00028
• 发表时间: 2023-05
• 期刊: 2023 IEEE International Parallel and Distributed Processing Symposium (IPDPS)
• 作者: Di Zhang;Chris Egersdoerfer;Tabassum Mahmud;Mai Zheng;Dong Dai

A Shared Memory Cache Layer across Multiple Executors in Apache Spark

Apache Spark 中跨多个执行器的共享内存缓存层

• DOI: 10.1109/bigdata50022.2020.9378179
• 发表时间: 2020
• 期刊: 2020 IEEE International Conference on Big Data (Big Data
• 作者: Rang, Wei;Yang, Donglin;Cheng, Dazhao
• 通讯作者: Cheng, Dazhao