Multi-threading support for system-level performance simulation of multi-core architectures

Abstract:
Multi-threaded programming is gaining popularity as general purpose processors have evolved to multi-core platforms. This brings new challenges for software designers. In the early phases of multi-threaded applications development designers have to do design decisions regarding load-balancing, thread management and synchronization. Even for moderately complex applications with relatively small number of concurrent threads, the design space will be large and its exploration requires that the designer is able to quickly evaluate performance of the candidate software architecture on the chosen platform. System-level performance simulation of the applications and platforms using abstracted workload and processing capability models suits to this purpose. These virtual system models enable fast simulation of multi-threaded applications in the early phases of the design process with reasonable modeling effort. This paper presents a performance modelling and simulation approach to explore efficiently multi-threaded applications and multi-core architectures at system-level , while exploiting earlier developed overall approach. Abstract workload models are generated from POSIX threaded application source code and mapped onto execution platform for transaction level simulation. The resulting workload models are simulated using network traffic monitor application example.