CS595-03: OS and Runtime System Design for Supercomputing

Overview

When one writes programs that push the limits of the hardware, the general abstractions and convenient programming environments provided by typical operating systems can actually block progress towards achieving breakneck performance. Computer systems are ultimately driven by their usage, and OS designers can hardly be expected to optimize for every use case. In high-performance and parallel computing, where no one use case can be described as the "common" one, we must consider specialization of our systems software. We must sometimes choose simplicity and speed over generality and elegance. With this theme in mind, we will approach various topics in the design of operating systems and runtime systems as they apply to supercomputing.

This is a seminar course that will expose you to current research at the intersection of several areas within computer systems, including Operating Systems, Computer Architecture, High-performance Computing, Parallel Programming, and Programming Languages. Students will apply what they learn to complete a course project in the design of OS internals for a specific runtime system, parallel language, or parallel algorithm.

Audience

This course is aimed at advanced CS undergraduates and graduate students, particularly Ph.D. students. Please contact me if you would like to take this class but are not sure if you have the required background.

Prerequisites

I expect you to, at a minimum, be well-versed in programming in C in a UNIX environment and have basic familiarity with x86 systems. This essentially means you have taken CS 351 (Systems Programming) or an equivalent course. You should also have a firm grasp on basic operating systems concepts, to the level of CS 450, CS 550, or CS 551. Experience with Computer Architecture (CS 470), Parallel Systems (CS 546), or Data-Intensive Computing (CS 554) is a plus.

Books and Software

There are no required textbooks for this course. This means it is absolutely essential that you attend class, participate in discussions, and come to office hours. We will primarily be reading research papers. However, here are some books I recommend you have on your shelf for reference, especially if this is a topic you intend to pursue at a deeper level:

• R. Bryant, D. O'Hallaron, Computer Systems: A Programmer's Perspective (3rd Edition), Addison Wesley, 2015. (book site) (Amazon)
• A. Silberschatz, P. Galbin, G. Gagne, Operating Systems Concepts (9th Edition), Wiley, 2012. (Amazon)
• T. Anderson, M. Dahlin, Operating Systems: Principles and Practice (2nd Edition), Recursive Books, 2014. (Amazon)
• J. Hennessy, D. Patterson, Computer Architecture: A Quantitative Approach (4th Edition), Morgan Kaufmann, 2006. (Amazon)
• M. Hill, Readings in Computer Architecture (1st Edition), Morgan Kaufmann, 1999. (Amazon)
• M. Quinn, Parallel Programming in C with MPI and OpenMP (1st Edition), McGraw-Hill, 2003. (Amazon)
• M. Harchol-Balter, Performance Modeling and Design of Computer Systems: Queueing Theory in Action (1st Edition), Cambridge University Press, 2013. (Amazon)

I expect we will be using some subset of the following software:

• Nautilus Aerokernel
• Kitten Light-weight Kernel
• Palacios VMM
• Legion Runtime System
• Charm++
• UPC

Development Environment

We will primarily be using virtualization for these projects. That means QEMU, QEMU+kvm, or Palacios. The setup is fairly easy and I will go through it either in class or on Piazza.

Grading

The following components will constitute your grade in this course:

• In-class discussion: 30%
• Project: 50%
• Project paper and presentation: 20%

Project

You will work in groups on a research-oriented project in OS or runtime systems. I will hand out a list of potential projects at the beginning of the semester. Once you have formed groups (I encourage you to use Piazza for this if you cannot find someone in person) and selected a project, I will expect weekly status reports from you until the end of the quarter, at which point you will submit a final report and give a final project presentation that includes a demo of your work.

Communication

We will be using Piazza for discussion and announcements. I will enroll you in the course page. I suggest you post your questions and comments on Piazza first before using direct e-mail. This way, the whole class can benefit from the answers. I will also use Piazza to post details on readings, background information, and other miscellaneous advice.

Schedule

What You Will Learn

• How to read and critique research papers in systems, particularly in OS
• How to present work in an academic setting
• Historical perspectives on OS and supercomputing
• Historical Perspectives on Supercomputing
• Navigating, understanding, and improving research codebases
• OS concepts and programming techniques, particularly as they apply to HPC
• OS architectures, both historical and modern, for HPC
• Parallel languages and runtime systems
• Hardware architectures for HPC
• Various low-level systems issues

Other Useful Links and Resources

• Mike Dahlin's Advice to system's researchers
• Writing a Good Systems Paper
• How to Have a Bad Research Career in Academia
• Stanford Systems Qualifier reading list
• C. Murray, The Supermen: The Story of Seymour Cray and the Technical Wizards Behind the Supercomputer (1st Edition), Wiley, 1997. (Amazon)
• Butler Lampson: Computer Systems Research: Past and Future
• Richard Feynman: There's Plenty of Room at the Bottom (video)
• Another perspective on computing from Feynman