Converting a Parallel Code to a Parallel High Performance Application
Staff - Faculty of Informatics
Start date: 4 November 2009
End date: 5 November 2009
The Faculty of Informatics is pleased to announce a seminar given by Sadaf Alam
DATE: Wednesday, November 4th, 2009
PLACE: USI Università della Svizzera italiana, room SI-006, Informatics Building (Via G. Buffi 13)
Although transforming a serial method or algorithm implementation to a parallel one by decomposing data and tasks and by exploiting concurrency in the given algorithm is a nontrivial task, improving performance and scaling of a first-cut implementation on a range of platforms could prove to be far more challenging. The complexity of emerging architectures with multiple cores, deep memory hierarchies and 100,000-ways parallelism on high-end supercomputing platforms offer both opportunities and challenges for not only achieving but also sustaining parallel efficiencies at scale. In this talk, I will introduce fundamental concepts of performance evaluation and analysis, taxonomy of parallel programming models and architectures, key terminologies and demonstrate how underlying computer architecture and supporting system software including compilers could influence achievable performance of numerical kernels. Using a production-level application example on the CSCS Cray XT5 platform, I will also illustrate how performance tools could enable end users in identifying and locating scaling bottlenecks. The talk will conclude with some examples of synthetic benchmarks that characterize computation, communication, memory and file I/O subsystems performance on Petascale systems.
Sadaf Alam is a member of Scientific Computing group at the Swiss National Supercomputing Center (CSCS) in Manno. Before joining CSCS, Dr Alam was a computer scientist in the Future Technologies group and a member of Scientific Computing group at the National Center for Computational Sciences, Oak Ridge National Laboratory, USA. Her research interests include programming languages and models, and srchitecture for emerging high performance computing (HPC) systems and performance studies of supercomputing platforms and scientific HPC applications. She has published and has served as a program committee member and reviewer for international computer and computational science conferences and journals. She earned her PhD in computer science degree from the University of Edinburgh, United Kingdom in 2004. She is a member of IEEE and IEEE Women in Engineering.
HOST: Prof. Rolf Krause