Large-scale Low Latency Switch Design

Large-scale switch is one of the most important equipments in communication networks. In recent years, with the increase of communication service, the wide spread of fiber optic networks and the rapid development of cloud computing, the demand of the size and forwarding capability of current switching systems are growing very fast. Core switches in the communication network is under great pressure. In these circumstances, the study on large-scale switch has become one of the hot and promising topics today. In this paper, we will discuss this topic in the following three points.The first topic is the delay performance of feedback-based switch. The feedback-based two-stage switch is scalable as it is configured by a predetermined and periodic joint sequence of con-figurations. Its major problem is that the average packet delay is high under light traffic load. In this paper, we improve the performance of feedback-based switch while still ensuring in-order packet delivery and close to100%throughput. We first show that the different sequences of con-figurations may endow a feedback-based switch with different delay performance. We propose to devise a tailor-made sequence of configurations for the estimated traffic pattern. The optimal joint sequences that can produce the lowest average packet delay is formulated as an ILP (Integer Linear Programming) problem. The simulation results demonstrate that the optimal joint sequence can cut down the average packet delay up to50%. Even under random uniform traffic,14%performance improvement can be obtained. Last but not least, we also design a fast suboptimal algorithm for the practical implementation.The second one is the large-scale switch network topology design. The network topology plays an important role in switch routing method, fault tolerance, power-consuming and ability to resist network attack. We analyze the properties and drawbacks of the popular fat-tree. Then we propose a flexible link-grouped fat-tree. We analyze its properties and prove that it has full bisection bandwidth. At last, we propose to devise a link-grouped fat-tree with low packet delay and least SEs. Finally, we study the flow control of linecards in large-scale switch network. Flow control, which is the key difference between different system interconnects, can make the delay of packet lower. We propose a lossless flow control based on prediction and the request-grant system to make packets go through the switch network without unnecessary delay.