A speculation-based approach for performance and dependability analysis: a case study

In this paper, we propose two speculation-based methods for fast and accurate simulation-based performance and dependability analysis of complex systems, incorporating detailed simulation of system components. The first approach applies to performance analysis and the second to dependability analysis. Our target example is a networked cluster with compute nodes and one I/O node. Detailed simulation of the cache subsystem of the I/O node is conducted, and more abstract simulation of the compute nodes and the switching network is performed. Performance measures obtained include cache miss ratio and cache subsystem access time. Dependability measures obtained include error coverage of EDAC code and error detection latency distribution of errors introduced to the cache components. The two methods are implemented on a network of workstations