Description: The early decades of research on computer system design and operation have been marked by an utmost emphasis on performance metrics such as throughput and number of instructions completed per unit time. More recently, we have been witnessing an increasing interest in sustainable computing research and practice, where power, energy, and thermal management are considered as first-class system design and operation objectives.
This paradigm shift is rooted in multiple developments in recent past. First, with the advent of battery-powered embedded computing and mobile, ad-hoc, and sensor networks, Power/Energy has been recognized as a precious system resource that must be managed dynamically and optimally. This often involves re-visiting and fine-tuning several key system parameters such as processor speed, scheduling policy, memory/disk management algorithm, wireless communication range selection, while maintaining traditional system performance guarantees (e.g., timeliness, connectivity, output quality). In addition, power and thermal management is now considered a must for high-end systems such as servers and Internet Data Centers directly connected to the power grid. Internet Data Centers, such as those operated by the industry giants Google, Facebook, and Amazon, are the backbone of our current IT infrastructure. Google alone is running more than 1,000,000 servers consuming more than 260 million watts. In other words, Sustainable Computers and Networks are the key technologies for Cloud Computing.
Hewlett-Packard, Intel, Microsoft, Phoenix, and Toshiba have developed an open industry standard (ACPI) for power management on desktops, servers and laptops proving the equally pressing priority of the area for the industry. In fact, the primary trigger behind Intel's decision to shelve the initial plans for faster but power-hungry chip designs in favor of the multi-core architectures was the energy efficiency of the latter. Despite the fact that literally hundreds of papers have been published in the area in the last decade, some problems of great theoretical and practical significance remain still open. The International Technology Roadmap for Semiconductors has listed Power Management as well as Performance and Power Dissipation in High-Performance Applications as two near-term grand challenges.
This is a seminar-type course with strong emphasis on hot research issues in system-level power, energy, and thermal management. As such, the focus will be on operating systems-, network- and application-level techniques to manage power, energy, and temperature. Most of the existing techniques aim at managing power at the cost of degraded operation (e.g. reduced throughput, speed or transmission range) yielding novel and intriguing Computer Science Problems: how to keep the network connected with reduced transmission ranges, how to meet all the deadlines with reduced CPU speed, how to provide acceptable QoS guarantees for multimedia streams with reduced frame rates, how to satisfy the Service-Level-Agreements (SLAs) for servers while keeping most of the system components in idle states as much as possible, and many others.
During the term, we will present, discuss and evaluate various papers in Operating Systems, Real-Time Systems and Networking research on sustainable computing. Anotehr focus will be power, energy, and heat management in Internet Data Centers and Cloud Computing in general. Through a comprehensive term project, the students will be able to focus on a well-defined area of sustainable computing and perform a preliminary research. There will be no exams. The course is particularly suitable for PhD students and advanced MS students interested in hot research issues in the general Systems area. The course will also satisfy the breadth requirement of the MS in CS program for the Systems and Networks area. A tentative list of discussion topics include:
Prerequisites: CS 571 (Operating Systems) or permission of the instructor. The course is open to both Ph.D. students and advanced M.S. students.
Readings: There is no required textbook. Most of the course material will be provided by the instructor and through recent research articles.
Grading:
Presentations: During the first part of the course, the instructor will present the fundamentals of low-power computing and main research problems of the area. In the second part, the students will present articles from recent conference/workshop proceedings and journals. A list of suggested papers will be provided, however, the student suggestions are welcome. The (in-class) presentation will include a critical evaluation and discussion of the paper. The students will be required to read, and submit a brief summary/evaluation of the papers presented in class.
Term Project: Each student is expected to complete a term project and submit a research paper/report by the end of the term. Again, a list of potential projects will be provided; but students may define their own project as long as the project has sufficient scope/complexity and the instructor's approval is obtained. A term project may be in any of the following forms:
All students must abide by the GMU
Honor Code and CS
Department's Honor Code and Academic Integrity Policies during the
semester. We reserve the right to use automated tools such
as MOSS to detect plagiarism.
Violations of the Honor Code will result in an
F.
Class Home Page: Throughout the term, all course material (slides, handouts, readings, etc.) will be available on the GMU Blackboard system.
Disability Statement: If you have a learning or physical difference that may affect your academic work, you will need to furnish appropriate documentation to GMU Disability Resource Center. If you qualify for accommodation, the DRC staff will give you a form detailing appropriate accommodations for your instructor. If you have such a condition, you must talk to the instructor during the first week of the term about the issue.