Performance Impact of Object Oriented ProgrammingR

It is widely accepted that object-oriented design improves code reusability, facilitates code maintainability and enables higher levels of abstraction. Although the software engineering community has embraced object-oriented programming for these beneets, it has not been clear what performance overheads are associated with this programming paradigm. In this paper, we present some quantitative results based on the performance of a few programs in C and C++. Several programs were prooled and the statistics of several program executions at various compiler optimization levels were generated on two architectures, the MIPS and SPARC. One observation was that in spite of a static code increase in C++, the dynamic instruction counts were either comparable or smaller in C++. However the cache miss ratios and traac ratios were signiicantly worse for C++ (often twice). It was also seen that some of the C++ features such as function overloading and free unions did not incur any run time overhead. These results should be of interest to computer architects and compiler writers who are designing and optimizing systems for executing object-oriented programs. Although it is diicult to devise representative and comparable programs in two diierent languages that use two diierent programming paradigms, we believe that our study will provide some initial data points and hopefully will lead into more related research.