In this paper, we analyze several issues involved in developing low latency adaptive wormhole routing schemes for two-dimensional meshes. It is observed that along with adaptivity, balanced distribution of traffic has a significant impact on the system performance. Motivated by this observation, we develop a new fully adaptive routing algorithm called positive-first-negative-first for two-dimensional meshes. The algorithm uses only two virtual channels per physical channel creating two virtual networks. The messages are routed positive-first in one virtual network and negative-first in the other. Because of this combination, the algorithm distributes the system load uniformly throughout the network and is also fully adaptive. It is shown that the proposed algorithm results in providing better performance in terms of the average network latency and throughput when compared with the previously proposed routing algorithms.

In this paper, we present a new concept of region of adaptivity with respect to various routing algorithms in wormhole networks. Using this concept, we demonstrate that the previously proposed routing algorithms, though more adaptive, causes an uneven workload in the network which limits the performance improvement. A is observed that balanced distribution of traffic has greater impact on system performance than the adaptivity or efficiency of the algorithm. Based on these motivating factors, we have presented a new fully adaptive routing algorithm for 2-dimensional meshes using one extra virtual channel. The algorithm is more efficient in terms of the number of paths it offers between the source and the destination and also distributes the network load more evenly and symmetrically. The simulation results are presented and are compared with the results of previously proposed algorithms. It is shown that the proposed algorithm results in much better performance in terms of the average network latency and the throughput.>