Recently, adaptive routing algorithms which achieve a high performance compared with deterministic ones are commonly used in large scale parallel machines. Although their performance is influenced with an output selection function or output channel selection algorithm, only a few studies depending on specific topologies or specific traffic patterns have been done. In this paper, we propose two output selection functions called LDSF (load-dependent selection function) and LRU (Least Recently Used) selection function, which are not based on a specific topology nor traffic pattern. Result of simulations shows that proposed output selection functions improved the performance compared with traditional ones up to 30%.
展开▼
