next up previous contents
Next: 9.5 Application Design Concepts Up: 9. Smart Analysis Package Previous: 9.3 Applications


9.4 Performance Analysis

An analysis and comparison taking into account the performance of a whole-application are presented. It should be noted that for high precision simulations it is essential to model the simulation domain as exactly as possible. The accuracy and efficiency of a finite element or a finite volume simulation strongly depends on the quality of the tessellation of the domain. Figure 9.11 presents the structure under investigation with a coarse mesh for the following performance analysis. In order to obtain sufficiently accurate results, the mesh size typically has to be in the order of $ 10^4$ to some $ 10^5$ nodes.

Figure 9.11: Temperature distribution due to self-heating in a tapered interconnect line with cylindrical vias.
\begin{figure}\begin{center}
\epsfig{figure=figures/ic_1.eps, width=11.9cm}
\end{center}\end{figure}

Based on these observations of restrictions due to limited resources, the influence of a problem's size on the overall performance is also illustrated. The given test structure with different levels of mesh refinement is used for analysis. By resolving a three-dimensional simulation domain, the large number of points easily leads to severe problems caused by memory bandwidth restrictions (Figure 9.12).

Figure 9.12: Comparison of the finite element assembly times in equations per millisecond, GSSE and SAP.
\begin{figure}\begin{center}\small
\psfrag{1e+03} [c]{$\mathrm{10}^\mathrm{3}$...
...res/performance_gsse_sap.ps, width=9.5cm, angle=-90}
\end{center} \end{figure}

As can be seen, the high performance manually tuned SAP shows the same run-time compared to the GSSE code module.


next up previous contents
Next: 9.5 Application Design Concepts Up: 9. Smart Analysis Package Previous: 9.3 Applications

R. Heinzl: Concepts for Scientific Computing