CASPAR deployment in a multi-core environment Evaluation of impact on performance and area

The success of multi-processor systems depends to a large extent on the underlying memory coherence system. In large systems, this can quickly become exponentially complicated. This in turn leads to increasie in development as well as debugging time. As the size of the system increases so do the chances of bugs slipping through the debugging process and into the hands of consumers. One way to deal with such errors is to use in-field check-pointing schemes. This paper presents an implementation of CASPAR[3] which is a check-pointing mechanism. CASPAR allows patching of error patterns to detect and bypass errors in the cache coherence protocols. In order to evaluate performance, an experimental multi-core system was set up. CASPAR was introduced and the effects on normal performance were evaluated.