Hasso-Plattner-Institut
Prof. Dr. Tilmann Rabl
  
 

GPU-Accelerated Data Processing

Instructors

Prof. Dr. Tilmann Rabl , Ilin Tolovski

Description

As peak frequencies in the processors are reaching the maximum, many of the computationally heavy tasks are offloaded to other hardware components. The parallel architecture in the graphical processing units (GPUs) originally used for efficient rendering of scenes has been redesigned and specialized to accelerate data processing and data management tasks in database systems and machine learning. State-of-the-art research in GPU utilization goes in the direction of efficient workload distribution across multiple GPUs using fast interconnects between GPUs for local peer-to-peer communication, or InfiniBand for direct memory access between nodes in a cluster setup. These developments allow us to rethink and distribute fundamental operations (joining, aggregation, sorting) that were long tied to the CPU. In this course we will gain a deeper understanding of the developments in general purpose GPU (GPGPU) computation. Our aim is to introduce and evaluate new solutions for workloads that will utilize the most of the vast capability of modern GPU architectures. 

Structure

Paper presentations

In this course, the students will have the opportunity to prepare discussion sessions on the state-of-the-art research in data processing using hardware accelerators (GPUs). This includes studying a research paper in detail, presenting it in front of the group, introducing valuable insights, and leading the following discussion. These sessions will be held regularly (weekly or bi-weekly ) as the first part of the course after which the students will select their project topics. Depending on the COVID-19 situation, we'll meet either in person or online. 

Project

This seminar will result in a group research project using hardware accelerators (GPUs)  for data processing tasks. The students can work in pairs to develop their own research ideas, implement and evaluate them. At the end of the course, the students should hand-in a written report and give a final presentation on their project. If there is an interest, we'd proceed to publish the findings at a topic-related scientific conference.

Grading

  • Project + report - 50%
  • Final presentation - 20%
  • Paper discussions - 20%
  • Active participation - 10%

Announcements

  • Course management via Moodle. There we will make any announcements and share materials.
  • The course is limited to 4 students.
  • Kick-off meeting - 3rd November, 15:15 @ F-1.04 (and via online stream)
  • If you have any questions, please contact me at ilin.tolovski (at) hpi.de.

Schedule

Tuesdays - 15:15 @ F-1.04. All meetings will be also streamed online. In the meetings we will conduct the paper presentations or discuss the project progress.