A role for Pareto optimality in mining performance data

Abstract

SUMMARYImprovements in performance modeling and identification of computational regimes within software libraries is a critical first step in developing software libraries that are truly agile with respect to the application as well as to the hardware. It is shown here that Pareto ranking, a concept from multiobjective optimization, can be an effective tool for mining large performance datasets. The approach is illustrated using software performance data gathered using both the public domain LAPACK library and an asynchronous communication library based on IBM LAPI active message library.

Pattern Mining for Historical Data Analysis by Using MOEA

“…Several studies were surveyed for this categorization; however, most of them treated single objective optimization problems. Recently, some researchers have proposed multiobjective approaches for the problem presented in [3,6,8,9]. We believe that these are positive approaches because these data mining problems have various criteria to be fulfilled.…”