Statistically analyzing execution variance for soft real-time applications

Abstract. Certain high-performance applications like multimedia and gaming have performance requirements beyond reducing program execu-tion time. These applications have repetitive components whose desired performance characteristics are more naturally expressed using soft real-time theory with its probabilistic guarantees. However, for large com-plex gaming and multimedia applications, programmers typically avoid real-time constructs as they significantly constrain how the program-mer can express functionality. Instead, such applications are developed as monolithic programs using conventional languages like C/C++. Here the soft real-time behavior of the application becomes an emergent qual-ity rather than being enforced by design. Programmers must then tweak parameters/algorithms until the application’s soft real-time behavior be-comes acceptable. There are currently no analysis techniques that di-rectly extract the soft real-time execution characteristics of monolithic applications written without the use of real-time constructs. We intro-duce a domain-agnostic profiling methodology that identifies program execution-contexts whose variant behavior most significantly affects the soft real-time characteristics of the application.