2007年8月龙星课程 http://dragonstar.ict.ac.cn/ 周源源老师课程体会 包云岗 中科院计算所 2007.8

Department of Computer Science Slide 2 周源源是谁？ UIUC的副教授 经历 2000年毕业于普林斯顿，导师李凯 1996之前在北大读本科与硕士 1999~2002年在NEC，StorageNetworks等公司就职 2001年到UIUC从教，并于2005年获得终生教授 获奖 UIUC Gear Faculty Award (2006) CRA-W Anita Borg Early Career Award (2005) IBM Faculty Award (2004 and 2005) DOE Early Career Award (2005) IBM Shared University Research (SUR) Award (2003) NSF CAREER Award (2004) YY. Zhou Department of Computer Science Slide 2

Opera：Operating Systems Research on Energy, Reliability and Autonomy YY. Zhou Department of Computer Science Slide 3

Opera：Selected Publications 2002 1 ISCA 2003 1 SIGMETRACS 1 USENIX 2004 1 HPCA (17%) 1 FAST (20%) 1 USENIX (13%) 1 ISCA (14%) 1 ICS (20%) 2 ASPLOS (14%) 3 Micro (18%), 2 Top Picks 1 OSDI (14%) 2005 1 HPCA (15%) 1 SIGMETRICS(13%) 1 HOTOS 2 SOSP (13%), 1 Awarded 1 Micro 2006 1 HotDep (15%) 1 ASPLOS (22%) 2 Micro (22%) 2007 1 HPCA 1 Micro, Top Picks 1 HotOS 1 HotDep 1 SIGMETRICS 3 SOSP YY. Zhou Department of Computer Science Slide 4

Department of Computer Science Slide 5 Opera：巧 选择的方向非常巧 分析系统的缺陷，Bugs，Errors 找到就是成功 使用的方法非常巧 统计 数据挖掘 自然语言处理 写文章技术非常巧 能把握Reviewer所关心的内容 强调某一点以得到至少一个PC的支持  Opera是一个成功的研究团队 YY. Zhou Department of Computer Science Slide 5

Department of Computer Science Slide 6 分享几点体会 研究方法 团队管理 学生培养 YY. Zhou Department of Computer Science Slide 6

研究方法: Problem-Driven Research Process Phase 0: Research direction Identification It is an ART Ask, Read, & Think 关注点 Important Interesting Background YY. Zhou Department of Computer Science Slide 7

Problem-Driven Research Process(1) Phase 1: Problem space understanding(1-2 months) From high level to low level Identify a concrete, important, open problem Problem may also come out from your previous experience and insight Literature study to get the state of the art to make sure it is open Search and feel around to see if it is important Concrete examples examination YY. Zhou Department of Computer Science Slide 8

Problem-Driven Research Process(2) Phase 2: Solution Seeking (1-2 months or more) Think, Brainstorm, Try Phase 3: Feasibility Study (1~2 months) Fast Prototype Simple Statistic Collection Manual idea walk through concrete examples Phase 4: If feasible, thoughts synchronization (2-4 weeks) Write introduction and related work Determine selling points Determine major research challenges YY. Zhou Department of Computer Science Slide 9

Problem-Driven Research Process(3) Phase 5: System and experiment design (2-3 weeks) Put design and experiment design in paper Draw empty figures in paper Determine which system to build on & which applications to use Phase 6: Implementation (3-6 months) Coding, debugging, testing, write impl. section May need to back to phase 3~5 to adjust design Changes are always recorded in writing YY. Zhou Department of Computer Science Slide 10

Problem-Driven Research Process(4) Phase 7: Result Collection First need to know what result we shall expect System tuning Fill in results in paper Phase 8: Proof read papers Ask others for feedback Wait for 1 week and read the paper again Don’t EVER argue with people who give feedback If the don’t understand it, it is YOUR fault You are not there to argue with reviewers and readers. YY. Zhou Department of Computer Science Slide 11

Department of Computer Science Slide 12 几个建议 如何很好的提出idea 先在Workshop上发表short paper提出idea 然后在随后的会议上发表full paper 多去公司实习 到公司，如IBM、Microsoft、Google等获取确切问题 每年暑假都要派学生去公司实习 建立学术关系 不要在意论文的署名次序 高年级带低年级学生  毕业后同门是最好的关系网 不要在意多加牛人 他们的推荐给学生的工作有很大的帮助 YY. Zhou Department of Computer Science Slide 12

Department of Computer Science Slide 13 研究方法总结 寻找钉子(具体问题) 选择合适的锤子 系统分析 统计 机器学习 制造强大的锤子 寻找钉子 YY. Zhou Department of Computer Science Slide 13

Department of Computer Science Slide 14 分享几点体会 研究方法 团队管理 学生培养 YY. Zhou Department of Computer Science Slide 14

团队管理 (Group Management) Reading Meeting Weekly Meeting Friday Ice Cream Topic Writing Book YY. Zhou Department of Computer Science Slide 15

Department of Computer Science Slide 16 Reading Meeting 选择2篇文章 跟踪会议大多数文章：ISCA，SOSP，OSDI， ASPLOS 选择部分会议相关文章：软件工程，网络等 报告 一个人报告 其他人模拟Program Committees (PC) 写Review Summary, Strength, Weakness, Borrow Ideas 给报告人打分 然后讨论文章内容 趣味讨论 Who is Top？-- 收集各个领域顶级的研究小组 研究内容，成员，主要成就…… Who is Who? -- 收集各个领域的牛人资料 研究方向，主要成就，毕业学校，导师与学生，照片 定期作个小测试，特别是开会前 YY. Zhou Department of Computer Science Slide 16

Department of Computer Science Slide 17 打分表 YY. Zhou Department of Computer Science Slide 17

Department of Computer Science Slide 18 Weekly Meeting 介绍本周工作情况 一定要带着问题来讨论 如果学生没有问题，不接受汇报，直接回去 周老师的观点 学生仅仅汇报作了什么  是为老师而做 学生带着问题来讨论  是为自己而做 YY. Zhou Department of Computer Science Slide 18

Department of Computer Science Slide 19 Friday Ice Cream Topic 每周五会有一个冰激凌沙龙 大家可以随便讨论问题 如果要好的idea，记录并分成几个subclass，以后几次冰激凌沙龙，针对各个子话题展开讨论 几次沙龙后，问题越来越具体，就会派学生着手调研，开始准备研究 花絮 开始的费用是罚款 如果有学生在小组内说普通话，就会被罚款 第一次5毛，第二次1元，第三次2元，…… 后来说的人少了，老师掏钱~~ YY. Zhou Department of Computer Science Slide 19

Department of Computer Science Slide 20 Writing Book 先写Introduction和Related Work，然后再设计实验 每篇文章的撰写从第一稿开始保留所有的颁本 第一稿发给小组内所有人，接受所有人的feedback，不允许反驳 修改后准备提交的颁布也要发给所有人，听取其他人的feedback 等1星期后再阅读自己的文章 等文章录用后，专门组织一次讨论 YY. Zhou Department of Computer Science Slide 20

Department of Computer Science Slide 21 分享几点体会 研究方法 团队管理 学生培养 YY. Zhou Department of Computer Science Slide 21

Department of Computer Science Slide 22 学生培养 招学生的两个标准 Ambitious, Self-Motivation Talk to Advisor 每个学生的培养 观察学生的特点 前两年在组内所有的项目中轮换 提一些开放性的问题，让学生去思考 1个高年级学生带2~3个低年级学生 高年级学生负责整体把握，工作协调，论文撰写 她的体会：导师就像刘备 自己不强 但是知道每个人的优点，会用人 她的学生都是5年毕业！ YY. Zhou Department of Computer Science Slide 22

Department of Computer Science Slide 23 我们可以借鉴什么？ 研究方法 多思考具体问题 把会议看作考试，以会议作为一些阶段目标 论文交流 模拟PC，写Review，打分 趣味讨论 写文章 先写Introduction与Related Work 整体构思  画空白图表  填数据 多宣传，多Show YY. Zhou Department of Computer Science Slide 23

Department of Computer Science Slide 24 体系结构领域相关资料 http://www.cs.wisc.edu/arch/www/ 主要会议 体系结构：ISCA，ASPLOS，MICRO，HPCA 高性能计算：PODC，PPoPP，SC，ICS，IPDPS，ICDCS 性能分析：SIGMETRICS 操作系统：OSDI，SOSP，Usenix 更多可以关注： http://libra.msra.cn/ YY. Zhou Department of Computer Science Slide 24

谢谢！ 欢迎大家讨论！