T e a P A C S
First International Workshop on Teaching Performance Analysis of Computer Systems
2021 is the 50th anniversary of ACM SIGMETRICS, while 2022 is the 50th anniversary of IFIP TC 7 (which includes WG 7.3 Computer System Modeling). In half a century, computing systems have changed drastically, and the growing complexity of systems, services and their relative context makes it more urgent and critical to model and analyze systems performance and behavior. On the other hand, the teaching of performance modeling has faced cutbacks, as budgets are squeezed and student interests shift. This is a good time to take stock of performance modeling as a discipline and this Workshop aims to discuss how our community should respond to this challenge.
For example, how should course content in performance analysis be updated for undergraduate and graduate students? Should we reach out to the industry, to import case studies, as well as educate engineers and managers? Can we demonstrate the relevance of performance analysis to other disciplines, and adopt pedagogical approaches from currently popular courses?
We issue here a call for academics, engineers and students to participate in this discussion at the workshop. We have invited 5 speakers who will provide keynotes that touch on the above issues. The success of the workshop, however, depends on a productive dialog and deliberation among participants, to summarize the state of our discipline, and suggest a course of action for growing performance modeling.
2.30pm-2.35pm Milan, 8.30am-8.35am NYC
2.35pm-3.05pm Milan, 8.35am-9.05am NYC
Mor Harchol-Balter: The most common queueing theory questions asked by computer systems practitioners
– Extended abstract
3.05pm-3.35pm Milan, 9.05am-9.35am NYC
Chee Wei Tan: The Value of Cooperation: From AIMD to Flipped Classroom Teaching
– Extended abstract
3.35pm-4.05pm Milan, 9.35am-10.05am NYC
Cathy Xia: Teaching Performance Modeling via Software and Instructional Technology
– Extended abstract
4.05pm-4.15pm Milan, 10.05am-10.15am NYC
4.15pm-4.30pm Milan, 10.15am-10.30am NYC
Vittoria de Nitto Personè: Teaching Performance Modelling 50 years later: Where Are We Going?
4.30pm-5.10pm Milan, 10.30am-11.10am NYC
Discussion 1: The current situation
5.10pm-5.40pm Milan, 11.10am-11.40am NYC
Jean-Yves Le Boudec: Performance Evaluation as Preparation for Statistics and Data Science
– Extended abstract
5.40pm-5.50pm Milan, 11.40am-11.50am NYC
5.50pm-6.20pm Milan, 11.50am-12.20am NYC
Giuseppe Serazzi: Updating the Content of Performance Analysis Textbooks
– Extended abstract
6.20pm-7.00pm Milan, 12.20am-1.00pm NYC
Discussion 2: What we can do
7.00pm-7.10pm Milan, 1.00pm-1.10pm NYC
“The most common queueing theory questions asked by computer systems practitioners“
Speaker: Mor Harchol-Balter, Carnegie Mellon University, USA
Abstract: As someone who has consulted with industry for many years, I find that there are many performance modeling questions which come up repeatedly. These include basic questions like: What is a “job?” What is load? What is throughput? What is an open system versus a closed-loop system? How can one understand one’s workload, with respect to mean, variability, tail behavior? They also include less basic questions like: What scheduling policy should I use to minimize mean response time? How about the tail of response time? If I favor short jobs, will I hurt the long ones? How can I schedule jobs if I don’t know their sizes? How should I handle jobs with different monetary values? What dispatching (load balancing) policies work in what settings? What if the jobs are multi-server (parallel) jobs? The purpose of this talk is to foster discussion about what performance modeling questions are most relevant to industry, along with a set of references (books, papers) that address those questions. Please bring your own favorite industry questions!
Short Bio: Mor Harchol-Balter, Bruce J. Nelson Professor of Computer Science, Carnegie Mellon University. Author of Performance Modeling & Design of Computer Systems: Queueing Theory in Action, Cambridge University Press, 2013.
“The Value of Cooperation: From AIMD to Flipped Classroom Teaching“
Speaker: Chee Wei Tan, City University of Hong Kong, China
Abstract: The well-known Additive Increase-Multiplicative Decrease (AIMD) abstraction for network congestion control was first published by Dah-Ming Chiu and Raj Jain in their seminal work in 1989 and soon played a prominent part in TCP algorithm design for the Internet. The ingenuity of AIMD lies in the abstraction of Internet congestion control, and ever since its inception has also been a staple part of teaching curriculum for performance evaluation and computer networking courses at universities worldwide. In this paper, we describe teaching examples for university students to appreciate the AIMD abstraction from the theoretical aspects such as convex optimization and Perron-Frobenius theory to the data science aspect. The essence of cooperation encompassed by AIMD reverberates even in teaching networks formed by students and educators, giving rise to online classroom flipping teaching tools and data analytics to close the gap between teachers and students.
Short Bio: Dr. Tan, Chee Wei received the M.A. and Ph.D. degrees from Princeton University. He was a Postdoctoral Scholar with Caltech, a Senior Fellow of the Institute for Pure and Applied Mathematics and a Visiting Faculty at Qualcomm and Tencent AI Lab. He is currently an associate professor in computer science at the City University of Hong Kong. His research interests include artificial intelligence, networks and graph analytics, online learning and convex optimization theory. He was a recipient of the Princeton University Wu Prize for Excellence, the Google Faculty Award, and several teaching awards. He is currently an IEEE Communications Society Distinguished Lecturer and has been an Associate Editor of IEEE Transactions on Communications and IEEE/ACM Transactions on Networking.
“Teaching Performance Modeling via Software and Instructional Technology”
Speaker: Cathy Xia, The Ohio State University, USA
Abstract: Performance modeling and analysis has become a common practice in the rapid development of modern information networks and service systems. The teaching of performance modeling today is faced with a number of new challenges: one should not only incorporate new topics to reflect the changing world ranging from information to economic to health crisis, but also embrace the proliferation of various forms of digital technologies in classroom teaching and learning. In this talk, the author will share her stories in teaching performance modeling utilizing software and digital technologies, with the purpose to foster further reflections and discussions.
Short Bio: Dr. Cathy H. Xia is an associate professor in the Department of Integrated Systems Engineering at the Ohio State University, where she has taught multiple courses on performance modeling, simulation, and stochastic processes. Dr. Xia is co-editor of book Performance Modeling and Engineering, Springer 2008, and guest editor for Cloud Computing as a Service, a special issue for Service Science 2013.
“Teaching Performance Modeling 50 Years Later: Where Are We Going?”
Speaker: Vittoria de Nitto Personè, University of Rome “Tor Vergata”, Italy
“Performance Evaluation as Preparation for Statistics and Data Science”
Speaker: Jean-Yves Le Boudec, EPFL, Switzerland
Abstract: After following a performance evaluation course, many students find that they are better equipped to fully grasp the meaning of statistics as used in data science courses. Perhaps this is because statistic is a branch of science that appears complex to many students, as it involves an in-depth understanding of what probability really means. Performance evaluation courses do exercise the theory of probability in powerful ways that can facilitate such an understanding. For example, writing a simulation program is a common exercise in such a course, and classical statistics can be well described by using the language of simulators. In turn, this can help understand the implications of residual scores in machine learning tasks. Another example is Palm calculus, which is at the heart of classical queuing theory and can be used to provide important insights into sampling problems encountered in data collection tasks.
Short Bio: Jean-Yves Le Boudec is professor at EPFL and fellow of the IEEE. He graduated from Ecole Normale Supérieure de Saint-Cloud, Paris, where he obtained the Agrégation in Mathematics in 1980 and received his doctorate in 1984 from the University of Rennes, France. From 1984 to 1987 he was with INSA/IRISA, Rennes. In 1987 he joined Bell Northern Research, Ottawa, Canada, as a member of scientific staff in the Network and Product Traffic Design Department. In 1988, he joined the IBM Zurich Research Laboratory where he was manager of the Customer Premises Network Department. In 1994 he became associate professor at EPFL. His interests are in the performance and architecture of communication systems and smart grids. He co-authored a book on network calculus, which serves as a foundation for deterministic networking, an introductory textbook on Information Sciences, and is the author of the book Performance Evaluation.
“Updating the Content of Performance Analysis Textbooks”
Speaker: Giuseppe Serazzi, Politecnico di Milano, Italy
Abstract: I have taught Performance Analysis courses for undergrad/grad students in universities with different fields of study: Mathematics, Informatics, and Computer Science. I have seen that the textbooks used in the courses change drastically depending on the field of study of the department. Other factors that I found have a strong influence on the lifespan of the textbooks are the speed of evolution of the techniques described, the changes in the architectures of the systems and services analyzed, and the personality of the authors. The first two are related to the characteristics of the technological trends, while the third one depends on the human characteristics of the authors. Indeed, some authors to demonstrate their profound knowledge often describe concepts with many unnecessary mathematical details, which generally create a fog shield that hides the key features of the concepts analyzed. The result is that the complexity of understanding the text is artificially increased and some students drop out. In this talk I will analyze the impact that deep changes in performance evaluation techniques, digital infrastructures and services have had on Performance Analysis textbooks to meet both academic and industrial needs.
Short Bio: Giuseppe Serazzi, professor emeritus, Politecnico di Milano, Italy. He taught Performance Evaluation courses at the University of Pavia, the State University of Milan, and the Politecnico di Milano.