CSR: Small: Improved Memory Management for Object-Oriented Big Data Systems

CSR：小：改进面向对象大数据系统的内存管理

• 批准号: 1907352
• 负责人: Harry Xu
• 金额: $ 27.06万
• 依托单位: University of California-Los Angeles
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2018
• 资助国家: 美国
• 起止时间: 2018-10-01 至 2020-07-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1907352&HistoricalAwards=false
• 关键词: CSR; Small; Improved; Memory; Management

This project seeks to develop runtime system support for improving the performance and scalability of a wide verity of data-intensive systems written in managed, object-oriented languages. It is clear that Big Data analytics has become a key component of modern computing. Popular data processing frameworks such as Hadoop, Spark, Naiad, or Hyracks are all developed in managed languages, such as Java, C#, or Scala, primarily due to the fast development cycles enabled by these languages and their abundance of library suites and community support. However, a great deal of evidence shows that memory management in Big Data systems is prohibitively expensive, severely damaging system performance. This project develops a series of runtime techniques that can automatically reduce the temporal and spatial costs of the managed runtime, allowing Big Data developers to fully enjoy the simplicity of managed languages without having to pay the performance price. Modern life is relying increasingly on Big Data analytics designed to support many concurrent users and quickly answer their queries. Behind visible services are data-intensive computing systems that need to quickly find useful information from a sea of data records, and therefore, their performance is critically important to our daily lives. This project provides an immediate performance benefit for such data-intensive systems, leading to improved quality, usability, and user satisfaction. The educational component of this project includes creation of new course materials, recruitment of undergraduate students and students from under-represented groups, and education of local programmers on how to develop highly-efficient Big Data applications.

该项目旨在开发运行时系统支持，以提高用托管的、面向对象的语言编写的大量数据密集型系统的性能和可伸缩性。很明显，大数据分析已经成为现代计算的关键组成部分。流行的数据处理框架（如Hadoop、Spark、Naiad或hyrack）都是用托管语言（如Java、c#或Scala）开发的，主要是因为这些语言支持的快速开发周期以及它们丰富的库套件和社区支持。然而，大量证据表明，大数据系统中的内存管理非常昂贵，严重损害了系统性能。该项目开发了一系列运行时技术，可以自动降低托管运行时的时间和空间成本，使大数据开发人员能够充分享受托管语言的简单性，而无需付出性能代价。现代生活越来越依赖于大数据分析，旨在支持多个并发用户并快速回答他们的查询。可见服务的背后是数据密集型计算系统，需要从海量的数据记录中快速找到有用的信息，因此，它们的性能对我们的日常生活至关重要。该项目为此类数据密集型系统提供了直接的性能优势，从而提高了质量、可用性和用户满意度。这个项目的教育部分包括创建新的课程材料，招收本科生和来自弱势群体的学生，以及教育当地程序员如何开发高效的大数据应用。