CSE254: Queuing Theory in practice: optimizing the performance of computer systems
- Times: Mon, Wed, Fri: 10:00 - 10:50
- Location: CENTR 216
- Section No.: 861025
Performance tuning of computer systems is an increasing challenge. While in the past the main focus was on peak performance, current system aim to balance cost, energy, heat dissipation and performance. Doing so requires the use of tools from probability and statistics, specifically, of Queuing theory.
This course will be based on the book "Performance Modeling and Design of Computer Systems / Queuing Theory in Practice" by Mor Harchol Balter.
The emphasis of the course is on the mathematical foundations of Queuing theory and the application to problems in computer resource allocation.
The connection of this course to machine learning can be described as follows: When managing a computer cluster, or computing in the cloud, we need to figure out how to provision compute resources. Queuing theory provides a well founded mathematical way to model how such systems work and what is the best way to provision given some parameters of the system and of the load.
In reality, one has little a-priori knowledge about these parameters. A promising direction is to combine queuing theory and machine learning to design computer clusters that adapt to the loads put on them and provide information about the current bottlenecks in the system.
The course requires a solid foundation in Probability and statistics. Specifically, students have to master Section II of the book "Necessary Probability Background" before starting the course. The first meeting of the class will be an entrance exam.
- Overview slides
- Links to papers contains links to papers by Mor Harchol Balter. The papers at the top of the page are the ones that have the closest relation to adaptivity and machine learning.
- presentations spread sheet Now with pointers to presentations (although some still missing)
- Queuing Theory Class Plan
- Piazza sign-up