부산대학교 도서관

In contemporary operating systems (OS), threads serve as the primary units of execution. Therefore, a comprehensive understanding of thread interaction is essential for system optimization. The Thread Interaction Graph (TIG) holds potential for analyzing dynamic systems that are sensitive to external stimuli and exhibit inherent unpredictability, characteristics often found in mobile devices. Despite its potential, TIG has not been extensively utilized. The existing body of research on TIG has been predominantly conducted under the presupposition of a static system, thereby excluding dynamic thread interactions. Our proposed method categorizes high-level interaction threads as major threads by utilizing thread interaction statistics obtained from TIG. Furthermore, our method delineates a group of major threads based on their relational attributes and formulates an interaction topology specific to this identified major thread group. In our experiments, we applied a thread scheduling constraint considering thread usage and cache efficiency, leveraging the interaction topology of the major thread group derived from our proposed method. This resulted in an 8% improvement in architectural efficiency compared to conventional scheduling methods.