A fast routing and bandwidth management method in ATM networks

Current supercomputers and workstations operate at 100 GFLOPS-TFLOPS and at 10-100 MFLOPS, respectively, and these processor speeds will increase in the future. This has led to an increasing demand for the real-time transfer of scientific high-dimensional image data, and ATM networks would therefore have to provide user-network interfaces with bandwidths in the gigabit range from now on. Because computer communication usually produces unpredictable burst traffic, it is difficult for the bandwidth reservation method based on statistical multiplexing to use bandwidths efficiently. If the connectionless server approach were adopted, these servers would cause throughput bottlenecks in gigabit applications. As a result, this paper proposes a fast routing and bandwidth management method that can effectively transport gigabit traffic. It also undertakes performance evaluation by comparing the proposed method with the P-NNI method that the ATM Forum proposes.