The documents distributed by this server have been provided by the contributing authors as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.
Publications of SPCL
|Model-Driven, Performance-Centric HPC Software and System Design and Optimization|
(Presentation - In , presented in Juelich, Germany, , ISSN: , ISBN: , Apr. 2011, Talk at Juelich Supercomputing Center (JSC) )
AbstractThe design and optimization of HPC software, middleware, and computing systems is a complex task. Two typical programming styles are (1) to tune a code until it is "fast enough" and (2) to tune a code until it is within "x%" of the peak performance. Method (1) is most common because there often exists no clear definition of "peak performance". Our goal is to integrate performance-centric design and analytic performance modeling into the application development process from the initial design phase through the application lifetime in order to assess the possible peak performance and move towards method (2). Analytic performance modeling can also be applied to optimize the middleware and system layers holistically based in the set of requirements of key applications. We will show manually developed performance models for key applications that will run at the full scale of the Blue Waters supercomputer system. We will discuss possibilities to support the modeling process in the middleware and programming model. We then outline several options to extend and adapt the middleware and system layers and the programming model to the application requirements based on analytic performance modeling.