Linear Network Coding Reduces Buffering in High-Speed Ethernet Parallel Transmission Systems

Abstract

Parallel transmission is a known technique of transmitting flows over multiple paths from a source towards the same destination. In high-speed Ethernet standards, for instance, large bandwidth flows are inverse-multiplexed into multiple lower-speed flows and transmitted in parallel. However, when flows traverse different paths, attention needs to be paid to the resulting differential delay, which requires computationally expensive path optimizations and large buffering at the receiver. In this paper, we show analytically that linear network coding can significantly reduce the buffering in high-speed Ethernet systems at a price of en/decoding overhead, while relaxing the requirements on path optimality. We implement the proposed decoding buffer model according to the IEEE 802.3ba standard, and show that linear network coding reduces the buffer size up to 40% as compared to systems without coding. With linear network coding, input interfaces of the destination node can deploy relatively smaller buffers, which is critical for wider practical deployment of high-speed Ethernet systems at 100 Gbps and beyond.