Abstract
In this paper, we present an Improved Balanced Adaptive Routing Protocol (IBARP) for low latency and high throughput.. By regrouping the cores around the congestion link and removing unnecessary groups in the original BARP, the traffic passes through the congestion link is reduced significantly. The congestion information can be passed to groups in the mesh structure directly that reduces hardware design complexity and lowers hardware cost. It is observed that IBARP reduces the average latency by 13-30% compared to BARP, static XY routing, oddeven routing and dynamic AD routing. ©2010 IEEE.