CAREER: High Performance Storage Systems

对网络服务的需求正以惊人的速度增长。这种不断增长的需求要求开发必要的基础设施，以促进对大量分布式信息存储的快速和轻松访问。本研究的重点是使分布式信息系统普遍有用所需的基本基础技术。具体而言，本研究包括：高性能文件系统设计与分析、旋转优化磁盘调度和分布式缓存数据。这项研究将产生高性能的分布式信息服务器，利用新的文件系统技术、先进的磁盘调度算法和基于网络拓扑位置的缓存（地理推送缓存）。文件系统的设计借鉴了当今日志结构和日志文件系统的基本特征，将它们与自适应的、旋转优化的磁盘调度技术相结合。为可伸缩性而设计的新缓存技术通过减少用户对分布式请求的感知延迟来增强文件系统的可行性。教学部分强调计算机科学专业和非专业的课程开发，以及本科生和研究生助教的培训计划。对于专业人士来说，专注于本科课程与哈佛建立一个强大的计算机系统研究小组的承诺相辅相成。本科课程已经并将继续修改，以纳入一个领域的基本概念，经典文献和研究活动。这种课堂活动的结合是培养能够充分参与正在进行的研究项目的本科生的基石。对于非集中课程的学生，计算机科学入门课程已被修改，以强调设计和解决问题的技能，而不是编程机制。随着设计入门课程的发展，未来的发展将进一步扩展这个想法。本课程旨在为计算机科学和工程专业的学生提供一个共同的基础，也为非专业的学生提供一个广泛的介绍。研究生课程开发的重点是科学方法在系统研究中的应用。学生被要求评论已发表的作品，设计实验评估平台，分析和评论数据集，并进行高质量的研究。教学建议的最后一个方面是开展助教研修班。萨尔茨教授一直与德里克·博克教学与学习中心密切合作，参加他们的讲习班，并与她的助教一起主持小组讨论和研讨会。她开发的入职培训程序将得到加强，使其适用于所有计算机科学课程的助教，并可能适用于其他领域。

The demand for network services is growing at an exceptional rate. This growing demand requires that the necessary infrastructure be developed to facilitate rapid and easy access to vast, distributed information stores. The focus of this research is the essential base technologies required to make distributed information systems generally useful. Specifically, this research includes: high performance file system design and analysis, rotationally optimal disk scheduling, and distributed caching data. The research will result in a high-performance, distributed information server, utilizing new file system techniques, advanced disk scheduling algorithms, and caching based upon network topological location (geographical push-caching). The file system design borrows the essential characteristics of today's log-structured and journaling file systems, combining them with adaptive, rotationally optimized disk scheduling techniques. New caching techniques designed for scalability enhance the file system's feasibility by reducing user-perceived latency for distributed requests. The teaching component emphasizes curriculum development for both computer science concentrators and non-concentrators and training programs for undergraduate and graduate teaching assistants. For concentrators, the focus on undergraduate curriculum complements Harvard's commitment to the establishment of a strong computer systems research group. Undergraduate courses have been and will continue to be modified to incorporate the fundamental concepts in an area, the classic literature, and research activities. This combination of class activities is the cornerstone for producing undergraduates who can fully participate in ongoing research projects. For non-concentrators, the introductory computer science sequence has been modified to emphasize design and problem solving skills over programming mechanics. Future developments will extend this idea still further with the development of an introductory design course. The course is intended to serve both as a common ground for computer science and engineering concentrators as well as a broad introduction for non-concentrators. The focus of the graduate curriculum development is on the application of the scientific method to systems research. Students are asked to critique published work, design experimental evaluation platforms, analyze and critique data sets, and produce quality research. The last aspect of the teaching proposal is the development of teaching assistant training seminars. Professor Seltzer has been working closely with the Derek Bok Center for Teaching and Learning, participating in their workshops and presenting panels and seminars with her teaching assistants. The orientation procedure she has developed will be enhanced so that it will be suitable for teaching assistants in all compute science classes, and potentially other fields.