Lib Metamorphosis: A Performance Analysis Framework for Exchanging Data Structures in Performance Sensitive Applications

Abstract

When software does not meet performance requirements, difficult decisions are made to change central data structures which may be costly financially and increase development time. In addition, monitoring how these data structures are used, and trying to understand performance implications of any change may prevent any evolution of the original infrastructure. Thus, radical revisions to software may be avoided due to the barriers of time and engineering complexity costs. Our solution to helping developers make infrastructure changes to improve performance is to provide a refactoring tool where developers may swap data structures. Our tool preserves correctness by utilizing the software's test suite and also measures performance automatically of the swapped data structure. We believe there is need for such a tool to help encourage more radical revisions and experimentation in large software projects to improve performance. Our frameworks success will be evaluated based on preserving the correctness of the software within a developer created test suite while providing performance information based on modified data structures.