The IBM Performance Simulator for Linux® on POWER® is a suite of performance models based on IBM's POWER series of processors. Users of Linux on POWER will be able to examine how their code executes on various IBM POWER processors so that they can identify and avoid common performance hazards on these processors.

The investigation is aided by a Java™-based graphical user interface (GUI) for viewing the flow of instructions through the processor models. This tool allows even inexperienced users to analyze their code through the aid of color-coding and real-time event definitions.

How does it work?

The performance models are software-based and trace-driven, which means that in order to run the model, one needs an instruction trace (an ordered list of the instructions executed) to drive the model. The models have no execution-driven mode and do not use data values. They produce only the number of cycles it takes to execute a given instruction sequence and, if directed to, a graphical representation of the internal instruction flow. The models are written by hand in a proprietary IBM language and verified against functional hardware models.

The main contributors to this tool are John Griswell and Douglas Logan. Other contributors include William Starke, J. D. Wellman, and Alex Mericas. All are employees of IBM.