Dr. Peter Tröger
Parallel Programming Concepts (2014)
Start: February 3rd 2014
Duration: 6 weeks
Course language: English
Since the very beginning of computer technology, processors have been built with ever-increasing clock frequencies and smarter optimizations for achieving a faster software execution. Developers and the software industry are used to applications becoming faster by merely exchanging the underlying hardware. However, since the beginning of the century it has become apparent that this approach no longer works.
Moore's law about the ever-increasing number of transistors per chip is still valid, but power consumption, thermal management and memory latency issues are making make serial code acceleration increasingly harder. Instead, hardware vendors now use additional transistors for multiple processing elements (‘cores’) per processor chip and deeper memory hierarchies. Modern hardware has the capability to transform any desktop, server, or even mobile system into some kind of parallel computer. This makes parallel programming the new default for application development. The exploitation of any additional horsepower from hardware is now in the responsibility of the software.
The openHPI online course “Parallel Programming Concepts” presents relevant theoretical and practical foundations for parallel programming. We show crucial theoretical ideas such as semaphores and actors, the architecture of modern parallel hardware, different programming models such as task parallelism, message passing and functional programming, and several patterns and best practices.
The course is suitable for all participants who are interested in getting a broader overview of parallelism, especially beyond the usage of multiple threads. Participants should have knowledge in at least one programming language - other skills are not necessary.
- Terminology and fundamental concepts
- Shared memory parallelism - the basics
- Shared memory parallelism - programming
- Distributed memory parallelism
- Patterns, best practices and examples
Dr. Peter Tröger
Dr. Peter Tröger is a senior researcher at the Hasso Plattner Institute for IT Systems Engineering at the University of Potsdam, Germany. He received a doctoral degree from University of Potsdam in 2008 for his work about stateful service-oriented infrastructures. Peter worked at the Blekinge Institute of Technology (Ronneby, Sweden), Humboldt University (Berlin, Germany) and the Brandenburg University of Technology (Cottbus, Germany), as researcher and lecturer. He is a member of GI and IEEE.
Peter’s current research interest is in dependability and programmability aspects of modern many-core environments. He is currently focusing on novel dependability modeling concepts and proactive failure prediction schemes. Peter acts as permanent program committee member for the DEPEND, ICSOC and SEKE conference series. He has research collaborations with the SAP Innovation Center, Univa, FZ Jülich, Audi, IBM Labs Böblingen, and Cloudera. Peter contributed to more than 40 publications and co-authored 3 books in the area of dependable parallel and distributed systems.
Peter currently leads the Distributed Resource Management Application API (DRMAA) working group at the Open Grid Forum. His work in this field resulted in a globally accepted standard for job management in large-scale parallel systems.
Frank Feinbube, M. Sc.
Frank Feinbube studied IT systems engineering at the Hasso Plattner Institute at the University of Potsdam where he graduated and received his master's degree. In 2009 Frank Feinbube joined the Operating Systems and Middleware group at Hasso Plattner Institute as a Ph.D. student and member of the HPI Research School on "Service-Oriented Systems Engineering".
Frank's research focus is on programming models for future hybrid computing systems. Current work is on managing complexity of hybrid parallel architectures (i.e.; GPU, Co-processors) with respect to ease-of-programming and resulting application performance. Further research interests include parallel programming languages and libraries, such as OpenCL, CUDA, OpenACC, OpenMP.