Praktikum: Fortgeschrittene Themen in Rechnerarchitektur und Parallelen Systemen (IN0012, IN2106, IN4242)

Vortragende/r (Mitwirkende/r)
Umfang6 SWS
SemesterWintersemester 2018/19
TermineSiehe TUMonline


Registration is done via the matching system.


  • Preliminary meeting: 22.06.2018, 14:00 in 01.06.020 (Slides)
  • Kick-off meeting: 18.10.2018, 16:00 in 01.06.020 (date changed on 2018-08-10!)
  • Intermediate talks: TBD (December/January 2019)
  • Final talks: TBD (likely March 2019)
  • Project report due: TBD (likely March 2019)

General Information

The participants work on current and advanced problems in the field of computer architecture and parallel systems. Students work in small groups of 2-3 people on a project. The results will be summarized in a final project report and presented to the other participants in an intermediate and a final talk.

(Tentative) Topics

Note: the following list is not complete and more topics can appear over time. Students are also encouraged to come up with their own ideas. The final assignment of topics and groups will be done at the kick-off meeting.

  • Tuning the LINPACK benchmark for the HimMUC cluster
    Goal is to maximize performance and energy efficiency of the HimMUC cluster on the High Performance LINPACK (HPL) benchmark and compare the results with other servers and architectures.
    Advisor: Alexis Engelke
  • Improved monitoring daemon and UI for the HimMUC cluster
    The current monitoring front-end is very simple - more information can be provided (e.g., statistics, or correlation to jobs) and the information can be shown in a more structured and visually appealing way.
    Advisor: Alexis Engelke
  • Fan control for the ODroid partition of the HimMUC cluster
    An adaptive fan control using an Arduino board can reduce the power consumption of the cluster by reducing the fan speed depending on the board temperatures. This is a technical topic and requires interaction with hardware.
    Advisor: Alexis Engelke
  • Installation and Testing of Open|SpeedShop on the HimMUC cluster
    Open|SpeedShop is a performance analysis tool, which we would also like to use on our HimMUC cluster. Task is the installation of the tool as well as to apply of it on one or two applications.
    Advisor: Alexis Engelke, Martin Schulz
  • Machine Learning for time-series analysis using HPC
    We are interested in deployment of Machine Learning models, namely auto encoders in HPC systems. In this project, first you get to know the LBANN, a Neural Network toolkit for HPC and use it for anomaly detection in time-series.
    Advisor: Amir Raoofy
  • Performance analysis and optimization of Matrix Profile algorithms
    Matrix Profile is a novel, efficient, data-mining approach to get an overview of time-series datasets which can be used for finding patterns and anomalies. For this topic, we will study performance of  implementations of algorithms for computing Matrix Profile on CPU and GPU and optimize them.
    Advisor: Amir Raoofy
  • Machine Learning based Strategy making for League of Legends
    Machine Learning is an emerging technology for data analytics and artificial intelligence. The task is to apply advanced machine learning techniques to analyze professional gameplay of League of Legends and find out a way to optimize.
    Note: This topic may also be completed as IN2016, IN2257 for Games Engineering students
    Advisor: Dai Yang
  • Compter Architecture for Space and Aviation Systems
    Space and Aviation systems often have strong requirements on their fault resilience, power budget, efficiency and real-time behaviour. These requirements opens new challenges for both Hardware and Software developer. We collaborate with Chair of Astronautics to provide Hand-on on real problems within research and engineering. 
    Advisor: Dai Yang
  • Your own idea in the field of computer architecture or parallel systems