The Technology of 25G, 50G, 100G

- Aug 26, 2019-

China's IDC circle reported on December 18 that the rise of cloud computing and the expansion of data centers drive the latest Ethernet speed upgrade, and big data based on cloud technology has also increased the workload of operators. To meet this need, data centers scale by adding bandwidth capabilities that are parallel to existing infrastructure. The expected rapid growth of 25G and 100G Ethernet deployments is evidence of this trend.

To handle the increasing data load, the industry's largest long-distance cloud companies are moving to 100G Ethernet architectures with their core network data center operators. But most carriers agree that 100 gigabytes or even 40 gigabytes is too much for a server connection, since their workloads need only gradually improve over a 10-gigabyte network. This is one reason why, despite the availability of 40G and 100G Ethernet, 25G and 50G Ethernet are still common choices within data centers. Here's why 25G is better for these applications than 40G.

25G SFP28 AOC .jpg



Several of the latest Ethernet bandwidth technologies were created not so much to achieve new speeds as to push the network protocol into neighboring markets, especially data centers. Below we will introduce 25G, 50G and 100G respectively to understand the specific reasons.

40G and 100G already exist, but why use 25G? This has puzzled some operators. The answer lies in architectural and performance requirements. The existing 100G standard network system consists of four links, each with a bandwidth of 25Gbps. This ratio of four to one is equivalent to connecting servers to a 25G switch, which then converges into a 100G uplink, helping network operators expand their data centres more easily.

Similarly, 40G Ethernet consists of four 10G Ethernet links. But according to John D 'ambrosia, President of the Ethernet Alliance, many data centres already have 10gb or more servers. This is why many chip manufacturers have offered 25G serial/unstring transceivers. Not only does this make it easier for 25G, 50G, and 100G Ethernet bandwidths to converge, but it also reduces cost by volume.


Although the IEEE standard for 50G Ethernet is still some way off (around 2018-2020), several industry consortia expect products to start appearing in 2016. Similar to 25G technology, 50G Ethernet technology will be the next solution for high-speed connections between servers and data centers. According to Dell 'oro, an analytics firm, servers and high-performance flash storage systems will need speeds of more than 25gb over the next few years.


SFP56G-AOC  hight speed cable.jpg

The 25G/50G Ethernet alliance has waived patent fees for the 25G and 50G Ethernet specifications and is open to all data center ecosystem vendors to help deliver these accelerated Ethernet technologies faster.

Reusing the 25G components of the existing 100G network reduces the implementation cost by 50G. For example, 25G cabling costs the same as 10G cabling but performs 2.5 times as well. Similarly, the 50G cost is half of the 40G cost, but the performance can be improved by 25%.


The deployment of 100G Ethernet will continue to grow for long-distance operator networks ranging from hundreds of kilometers to tens of thousands of kilometers.

100G QSFP28 TO 4SFP AOC.jpg



But according to one of the latest industry consortia, the 100G architecture is another excellent market alternative. According to the 100G CLR4 consortium, led by Intel and Arista Network, 100G is ideal for connecting large "very large scale" data centers ranging from 100m to 2km across.

Other companies are also looking at alternative 100G implementations for data centers. Avago Technologies has joined the CWDM4 MSA industry consortium, which aims to define a common specification for low-cost 100G optical interfaces within 2km of data center applications. As the network infrastructure is built to shift to a 100G data rate, data centers will require high-density 100G embedded optical connections over long distances. MSA provides four 25G single-mode optical fiber (SMF) link channels using coarse wavelength division multiplexing (CWDM) technology. Similarly, OpenOptics MSA, an organization sponsored by Ranovus and Mellanox Technologies, will focus its efforts on developing a data center capable of supporting 100G (2km) data centers.

In the past, it was speed that drove the development of most web components. Today, dealing with the vast amount of data streaming through the cloud requires companies to find a cost-effective solution that balances increased speed with reusability.