Performance Evaluation of Computer Systems and Networks

Teachers

• Prof. Giovanni Stea
• Ing. Giovanni Nardini

Course syllabus

• Probability theory and statistics (~30 h):
  • Fundamental definitions and theorems on probability. Uniform probability model. Discrete and continuous random variables. Notable RV distributions (exponential, uniform, Poissonian, normal, binomial, chi-square, student-t etc.). Central limit theorem.
  • Sample and population: estimators and confidence intervals. Data analysis and summarization. Model fitting, experiment design.
• Simulation (~30 h):
  • Principles of discrete event simulation: events, event queues, random number generation, structure of a simulation software.
  • Description of the general-purpose OMNeT++ simulation framework. Hands-on experiments with the OMNeT++ framework.
  • Simulation workflow: system modeling, experiment planning, factor reduction, independent replications, transient and steady-state behavior, output data analysis, experiment automation.
• Analytical System modeling (~30 h):
  • Queueing Theory: definition of stochastic process. Markov processes. Continuous-tmie Markov Chains.
  • Average measures: number of clients, waiting and queueing time. Little's theorem. PASTA theorem and the importance of the viewpoint.
  • Single-queue systems: M/M/1, M/M/c, M/M/1/k, discouraged arrivals, finite populations, batch arrivals and services.
  • Network of queues: open and closed networks. Jackson's networks. Gordon and Newell's networks. Buzen's algorithm.

Teaching material

Slides and exercises will be made available in the course channel on the Microsoft Teams platform.
The OMNeT++ Simulation manual can be found here.

Tools

In order to do exercises during labs and the group simulation project, students are required to install the OMNeT++ software on their computer (works on Windows, Linux and MacOS).
OMNeT++ can be downloaded from here. A detailed install guide is available here.

Available MSc Theses

• Digital Twins of Mobile Networks

A Digital Twin of a network can exploit simulation services for different purposes, e.g. what-if analysis, according to the "as-a-service" paradigm. This service can be instantiated on demand, possibly with constraints on the maximum simulation (wall clock) duration. Under such constraints, a simulator should be able to choose -- at run time -- the level of detail of the models that it runs, so as to produce the best possible results within the allocated time budget.
This can be done in several ways: running several (containerized) instances in parallel, and/or using multi-scale models of the same entity. Selection of the appropriate models should be performed based on performance profiling, possibly with the help of machine-learning algorithms.
This research will involve using and extending the open-source OMNeT++-based simulator Simu5G.

• MEC-based vehicular applications

Multi-access Edge Computing (MEC) will revolutionize the landscape of mobile application, by providing reliabe and ubiquitous computing power close to the edge of the network. This will allow one to run closed-loop control applications in real time, for instance cooperative, autonomous or teleoperated driving, robot coordination, etc.
The end-to-end performance of these services has not been sufficiently investigated yet. This is challenging, because they involve a number of potential sources of delay: onboard application, access (e.g., 5G) network, MEC computation, possibly MEC-based services.
Theses on this topic will have to propose creative solutions to problems like the above, and to evaluate them via simulation, using the open-source OMNeT++-based simulator Simu5G.