InfiniBand架构(IBA)是一种基于交换机的互连技术,拥有高带宽和低时延的特点。InfiniBand网络很适合构建高速网络集群系统并且已经被批准为I/O技术和进程间通信标准。提出了一种适用于电力广域高性能计算网络的基于动态工作阈值的有效流量控制方法。与以往的静态阈值方法不同,该方法根据一个周期内对链路中业务流量的监听,记录下突发大流量,最后根据所记录的流量特点来动态设定工作阈值。交换机对业务流量评估是一个动态连续的过程,对每一个业务流的最大突发值进行评估,动态调整仲裁表来设置恰当的阈值门限,使每个信道都能适应当前链路的流量特点,提高网络突发大流量的处理能力、带宽利用率和传输效率,降低网络能耗、丢包率和时延。采用最大熵原理(Maximum Entropy,ME)来分析所提出的流量控制机制,并用广义指数分布GE-Type模拟业务流到达率和业务服务时间。仿真结果表明,所提出的机制可以实现电力通信网络中广域高性能计算网络流量的有效控制。
InfiniBand Architecture (IBA) is defined as a switch-based interconnection technology with high bandwidth and low-latency, which is suitable for constructing high-speed networks for cluster systems and has been ratified as a new industry standard for the server I/O and inter-processor communication. An efficient flow control mechanism with dynamic job threshold for InfiniBand networks is propopsed. Unlike the existing static threshold methods,it dynamically sets threshold according to the recording traffic in a cycle of traffic monitoring on the link. Switch to traffic assessment is a dynamic and continuous process. The maximum burst value of each business flow to assess, dynamic adjustment of the table to set the appropriate threshold. It can improve the processing capacity, bandwidth utilization, transmission efficiency of network burst traffic and reduce the delay, blocking probability, mean queue length. The principle of Maximum Entropy (ME) is adopted as an effective methodology to analyze the new mechanism with the generalized exponential distribution( GE- Type) for modeling the inter-arrival times and service times of the input traffic. The simulation results show that it can achieve the effec- tive control of traffic flow in the high performance computing network in power communication.