Evaluation of highly conditionally diagnosable (n, k)-star topology for applications in resilient Network on Chip

摘要

With the rapid increase in the number of processor cores integrated on Network-on-Chips (NoCs) and higher requirements for system reliability, fault tolerance is becoming an important challenge in the design process. Some of research trends are focusing on system reconfiguration in situations with on-chip faulty cores. This is a relevant software research area that is an instance of a well known NP-complete problem. Instead of optimizing and analyzing the system in context of routing algorithms and applications, we found the topology itself can offer us with good properties such as fault tolerance and resilience. In this paper, we discuss the recently designed (n,k)-star NoC topology and examine its properties and advantages as compared to the popular 2-D mesh topology. Then, since the design of NoC topology requires addressing of issues and measurements specified by runtime evaluation observations, we introduce the metrics of performance (n,k)-star topology applied on NoCs. Related experiments and simulations are done as well to examine the performance of (n,k)-star topology and comparison to 2-D mesh topology. Our simulation demonstrates (n,k)-star topology outperforms same-scale 2-D mesh topology while exhibiting fault tolerance properties.