- Abstract:
-
Simulation has been established as the only cost-effective, efficient and flexible mechanism for conducting in-depth analyses of complex computer systems performance as well as to investigate novel and hypothetical system designs. For scientific supercomputing systems performance studies, system-specific and vendor-specific software prototypes are often developed at different design stages. However, there appear to be no simulation frameworks for performance evaluation and exploration of these tightly-integrated systems. This is because scientific supercomputing systems, unlike mainstream commercial clusters, often have unique design features and, in their life times, are dedicated to a few highly demanding applications.
The design and development of high-fidelity, scalable simulation models of high-end scientific systems is so far considered neither efficient nor cost-effective. In order to address these issues, we introduce the concept of simulation metamodelling; metamodelling allows for efficient generation of simulation models with alternative system configurations by maximising component reuse and minimising redesign overheads. The hardware-software co-simulation features of a computer architecture simulation framework, HASE, are exploited to build scalable, flexible and extensible simulation models. We compare and contrast our application-oriented simulation design approach with existing modelling practices, and subsequently identify its advantages and shortcomings.
- Links To Paper
- No links available
- Bibtex format
- @InProceedings{EDI-INF-RR-0528,
- author = {
S. Alam
and Roland Ibbett
},
- title = {A Methodology for Simulating Scientific Supercomputing Systems},
- book title = {Procs of Summer Computer Simulation Conference},
- year = 2004,
- month = {Jul},
- }
|