The Age of Ultra-Low Latency 25G Ethernet

The rapid evolution of networking technology has profoundly impacted various industries, enabling faster and more efficient communication and data processing. At the forefront of this revolution is 25G Ethernet, a technology that promises ultra-low latency and high performance for modern networks. This article explores the emergence of ultra-low latency 25G Ethernet, examining its evolution, needs, and current state.

In the early 2000s, achieving low latency was pivotal for electronic trading, leading to the development of various strategies to minimize delays. The introduction of high-performance 10G switches around 2008 marked a significant shift, reducing latency to hundreds of nanoseconds through cut-through architectures. Then the rush to switch to cut-through networking began. In order to guarantee that networks operated without interruptions from unintentional queuing or drops from congestion, these new networks made use of specialized switches with integrated traffic filtering and latency monitoring. While high-density systems with gradually reduced latency were introduced between 2008 and 2012, the electronic trading architecture—which made use of 10Gb leaf and spine systems for low latency trading—remained largely unchanged.

By 2013, Layer 1 switches further pushed the boundaries, delivering single-digit nanosecond latencies but with limited networking functions. These devices used FPGAs for essential packet processing, ensuring minimal delay in data transmission when one-to-many traffic is processed at the lowest feasible latency and just the necessary and minimal amount of packet processing is done via the FPGA for multiplexing as opposed to traditional switching.