Impact and utility of smell-driven performance tuning for end-user programmers

Abstract

This paper proposes a technique, called Smell-driven performance tuning (SDPT), which semi-automatically assists end-user programmers with fixing performance problems in visual dataflow programming languages. A within-subjects laboratory experiment showed SDPT increased end-user programmers’ success rate and decreased the time they required. Another study, based on using SDPT to analyze a corpus of example end-user programs, demonstrated that applying all available SDPT transformations achieved an execution time improvement of 42% and a memory usage improvement of 20%, comparable to improvements that expert programmers historically had manually achieved on the same programs. These results indicate that SDPT is an effective method for helping end-user programmers to fix performance problems.