Performance Evaluation
Professor | Serazzi Giuseppe |
Course program | MAS |
Year | 1 |
Semester | Spring |
Category | Elective |
ECTS | 3 |
Academic year | 2013/2014 |
Course type: Lecture
Value in ECTS: 3
Academic year 2013/2014 - Spring semester
Objective
As the complexity of investigated systems increases the performance forecast and evaluation is becoming more and more important and difficult at the same time. The course covers the basics of the main techniques needed to predict and optimize the performance of a computer system, a network or a application. The focus will be on simulation and analytical methods, probabilistic models, Markov and queuing network models. The concepts learned in the class will be applied in several case studies to evaluate and optimize the performance of centralised and distributed systems, multi-core and embedded systems, intranets, web servers, web-based services, cloud architectures, and wireless sensor networks. Optimisation techniques for maximizing the performance with multiclass workloads based on bottlenecks identification are also covered. Hands-on experience is provided through the use of the JMT suite of tools.
Contents
- The Operational Analysis technique (fundamental laws, operational equilibrium)
- Refresh on some basic probability distributions used in performance modelling
- Stations with queue (single and multiple servers, finite queue, general service times distribution)
- Asymptotic analysis (bounds and bottlenecks identification with multiclass workload)
- Single-class queuing networks (open, closed, MVA, approximate algorithms)
- Multiple-class queuing networks (MVA, Markov chain)
- Load balancing and performance optimisation (throughput maximisation, response time minimisation)
- Workload and traffic characterisation (burst analysis and simulation)
- Use of Java Modelling Tools JMT (http://jmt.sourceforge.net/) for performance evaluation studies
- Case studies (performance prediction of web services, capacity planning of web servers, wireless sensor networks, Intranet, modelling scheduling algorithms with Markov chains, load balancing and resource allocation to meet performance requirements, throughput maximisation through bottlenecks avoidance, ... )
Course material
Slides, articles, some chapters of specialised books, user manual of JMT tools