The Design And Implementation Of A Network-Coding Based Simulation Platform For Switches

Since the development of the Internet, multimedia, video conferences and other multicast applications become increasingly wide range, and people'requirements for the network quality of service (QoS) rapidly rise. But as the core of transmission network, the architectures of switching systems change several times, but still not fully meet the requirements. Combined Input and Output Queuing (CIOQ) may take advantage of Input Queuing (OQ) and Output Queuing (IQ) to some extent. However, even allowing fan-out splitting and inter-flow coding, as the switch size increases, there is not an upper limit of switching fabric speedup to achieve 100% throughput for multicast services. Graph theory model results show that network coding can increase up the achievable rate region for multicast, and can sustain traffic patterns that cannot be served if network coding were not allowed. To achieve 100% throughput, the upper limit for a minimum speedup reduces.Currently, the research of coding switching focuses on theoretical analysis, and there is some deficiency on applications and experiment data. According to the architecture of combination of parallel processing and crossbar fabric, with the CIOQ model, we design and implement the Network-Coding based Simulation Platform for Switches (NCSPS). NCSPS not only integrates the functions of traditional switching and coding switching, but also is convenient to implement new technologies with interfaces. NCSPS provides an important reference standard and comparison objects for switching research. There are mainly 4 features in NCSPS:(1) It is implemented in Java. NCSPS is object-oriented and provides cross-platform ability. In addition, it has high robustness and stability in multithreaded concurrency.(2) Each sub-module strives to"loosely coupled"in order to separate the traffic model, switching fabric and scheduling strategies. The sub-modules provide interfaces for each other, and cooperate driven by data flows.(3) It has good flexibility and scalability. Users can configure almost modules, and the interfaces are convenient to implement new technologies.(4) We focus on linear network-coding. The sub-modules only implement linear coding and decoding algorithms. According to experiment results, the linear coding and decoding algorithms meet the needs to a certain extent. With different configurations of input traffic model, port number, traffic load and scheduling strategy, we simulated switching systems with and without network coding, obtaining the results about cell delay, packet loss and queue share. Analysis on the simulation results shows that network coding switching achieves better performance on cell delay, packet loss and queue share, and the greater the load and the more the port number ,the more obvious the advantage is. When the load exceeds 0.8, the average cell delay of network-coding switching is 30% lower than that of traditional switching, and queue share and packet loss rate are 20% lower or more. This verifies the rationality of the coding theory, but also provides a new way of thinking and inspiration on the structure of switching systems.