1.3.1 Modeling of hardware software performance of high‐tech systems

Abstract

The performance of the control system is an important aspect of a machine. It would be a waste if a high-tech machine has been build such that it can physically achieve a high throughput, for example printed sheets of paper, but is limited because the software controlling it cannot keep up. Unfortunately, with current techniques it is hard to "predict" beforehand what the performance of the software will be when it finally runs in the real system on the real processor(s). There are two (extreme) ways to deal with it:1. Over-dimension the hardware platform to make sure the software will run. 2. Implement the software, then run and evaluate its performance on the target hardware platform. Then use this information in the next design cycle.The disadvantages of both approaches are clear. In the first situation the cost price of the entire system will surely be higher than necessary. In the second case, the design time is increased dramatically because more design cycles are needed. Therefore, it is important to strive to a development method that leads to fast design cycles for software performance, while having an accurate enough prediction. In this paper we will discuss a pragmatic modeling approach to design for performance in the domain of software intensive systems. Problem formulationAs explained, the goal is to find or develop methods, techniques and tools that make it possible to predict the performance of software accurately based on only a small model that does not need a lot work to come up with. Obviously, there is a tension between the accuracy of the performance prediction and the amount of work needed to make the model. In general, it is even likely (but not proven) that it will require more work to make an exact prediction of the performance than it would be to create the whole system, run it and see how it performs.During the process of performance modeling, and also during other Boderc activities, we realized that the goal of creating a model is not only to do an analysis and to make a prediction. Probably more important is the understanding that is obtained by creating the model, see also (Kostelijk 2005). This understanding leads to the ability to make better design choices and to be

Patterns for Automatic Generation of Soft Real-time System Models

“…These modeling patterns emerged from previous experience with modeling real-time systems (e.g. [31][32][33]).…”