The goal of this course, is giving the students the tools to predict and optimize the performances of modern computer architectures, that are characterized by a high level of parallelism, sharing and dynamic allocation of computational and storage resources, centralized and distributed components, exascale dimensions of resources and users, and a high variability of requests. To model such systems, new techniques have been introduced, and the existing ones have been enhanced. This course will present both analytical and simulative techniques currently used both in industrial and research environments. The course will be based on the "Learning through Applications" " approach, with which case studies in various domains will be analyzed. The will include e-health, private and public clouds, autoscaling data centers, energy consumption reduction, machine learning, smart cities and more.

Basic Knowledge of performance evaluation

Basic Knowledge of discrete event simulation

Basic Knowledge of performance measurement

Basic Knowledge of the Java Modelling Tool (JMT)

1. Introduction to performance indices and measures

The main performance indices that can be evaluated from a model, depending on the techniques being employed, will be presented. We will focus on response time, throughput, server utilization, confidence intervals, distributions, blocking probability, availability, reliability and more.

2. Workload and Service characterization

We will focus on the impact that different input and service processes can have on the system performances. We will focus on measuring an analyzing date taken from real systems, and how to mathematically characterize them with probability distributions, through fitting and generation of random traces starting from estimated data.

3. Analytical models

We will consider how to accurately model single service centers, considering multiple servers, finite capacity, and general service and inter-arrival time distributions.

4. Simulation and resource contention

Finally we will focus on simulation techniques, to study modern computing infrastructures. We will mainly focus on multi-class models to consider applications with heterogeneous workloads, Fork / Join and finite capacity systems to consider parallel processes, Petri Nets to model resource contentions, and dynamic algorithms to model adaptive techniques.

Probability and statistics

Linear algebra

The evaluation will be based on a project, that will be assigned to every students. It will consist in a pilot system (specifically created for the exam), that will have to be studied using the tools and techniques discussed during the course. At the end of each lesson, an assignment will be given to immediately put in practice what learned. The exam will then be an oral discussion of the project and of the assignments given throughout the course.

the first 7 chapters (free of copyright, from the course-site on Beep service)