Design and Evaluation of a Bio-Inspired, Distributed Middleware for a Multiple Mixed-Core System on Chip

This paper deals with the design and evaluation of a reliable middleware for multiple Mixed-Core Sytem-on-Chips. The middleware should thereby be responsible for the distribution of tasks among the processing elements. Additional to the requirement of a decentralized task allocation, the middleware should also use self-x principles of Organic Computing, like self-optimization or self-healing. Therefore, the here presented model utilizes another middleware, the artificial hormone system. Furthermore, the model introduces a cluster structure among the multiple Mixed-Core System-on-Chips. Among those clusters the set of tasks will be distributed and each cluster runs its own artificial hormone system with its subset of tasks. For distributing the tasks, the middleware uses a negotiation system, widely used by Multi-Agent-Systems. This hierarchical two-level concept brings also benefits when using it with a single Mixed-Core System-on-Chip. These benefits and the limitations of the system will be explained in this thesis. Also the results of the evaluation of the self-configuration and self-healing properties will be presented and discussed.