Performance Model for Parallel Mathematical Libraries Based on Historical Knowledgebase

摘要

Scientific and mathematical parallel libraries offer a high level of abstraction to programmers. However, it is still difficult to select the proper parameters and algorithms to maximize the application performance. This work proposes a performance model for dynamically adjusting applications written with the PETSc library. This model is based on historical performance information and data mining techniques. Finally, we demonstrate the validity of the proposed model through real experimentations.