Generation and calibration of compositional performance analysis models for multi-processor systems

Abstract

The performance analysis of heterogeneous multiprocessor systems is becoming increasingly difficult due to the steadily growing complexity of software and hardware components. To cope with these increasing requirements, analytic methods have been proposed. The automatic generation of analytic system models that faithfully represent real system implementations has received relatively little attention, however. In this paper, an approach is presented in which an analytic system model is automatically generated from the same specification that is also used for system synthesis. Analytic methods for performance analysis of a system can thus be seamlessly integrated into the multi-processor design flow which lays a sound foundation for designing systems with a predictable performance.

Mapping of applications to MPSoCs

“…The model preserves an accurate representation of the system, but it requires properly estimated parameters. The automated MPA model generation and calibration method is described in [9] and [10]. Figure 2 shows an MPA representation of a basic system with three processes mapped on two processors that communicate via a shared bus.…”

“…Ultimately, the same system-level specification of the application and architecture together with the optimal mapping specification form the synthesizable system specification, which will be implemented on the final system or can be simulated on the virtual platform. Typically, we use low level simulation in a feedback loop for automatically calibrating our time and thermal analysis models used for comparison of different mapping alternatives [9] [10] [34]. …”

Virtual prototyping for efficient multi-core ECU development of driver assistance systems

“…Another approach to calibrate a compositional performance analysis model can be found in [15]. To parameterize the model, input data is taken from hardware data sheets, system specification, low-level simulation and functional simulation.…”

Embedding formal performance analysis into the design cycle of MPSoCs for real-time streaming applications

“…This article is based on the work described in Haid et al [2009], in which synchronous dataflow [Lee and Messerschmitt 1987] was considered in which the number of tokens consumed and produced in each activation (firing) of an actor is constant. By using socalled consumption and production curves, this work is extended to dataflow process networks in which the number of tokens consumed and produced in each activation of an actor may vary due to data or state dependencies.…”