View the 25G, 50G, and G technologies in the data center

View the 25G, 50G, and G technologies in the data center

Enterprises and cloud-level network data centers have a variety of "G". It is time to figure out their performance.

 

The rise of cloud computing and the expansion of data centers have promoted the latest Ethernet Speed upgrade, while cloud-based big data has also increased the workload of operators. To meet this demand, the data center expands by adding bandwidth capabilities parallel to the existing infrastructure. This trend is evidenced by the anticipated rapid growth of 25g and G Ethernet deployment.

To handle increasing data loads, the industry's largest long-distance cloud companies have switched to the G Ethernet architecture together with their core network data center operators. However, most operators believe that G or even 40g is too high for server connections, because the workload only needs to be improved on 10G networks. This is one of the reasons why 25g and 50G Ethernet are common internal data centers, even though 40g and G Ethernet have been introduced. Next we will briefly explain why 25g is more suitable for these applications than 40g.

The formation of several newest Ethernet bandwidth technologies is not to set a new high speed, but more to push such network protocols into nearby markets, especially the data center market. The following describes the specific causes by introducing 25G, 50G, and 100G respectively.

25G

The official draft IEEE 802.3 standard for 25G Ethernet will be finalized in 2016 (see figure 1), primarily for cloud data center servers. Since 10g and G Ethernet components can be reused, this is a relatively short time frame.

 

Figure 1: Ethernet standard development plan

40G and G already exist, but why 25G? This puzzled some operators. The answer lies in the architecture and performance requirements. The existing GB standard network system consists of four links, each of which has a bandwidth of 25 Gbps. This 4-to-1 ratio is equivalent to connecting servers to a 25G switch and then aggregating them into a G uplink, which helps network operators expand their data centers more conveniently.

Similarly, 40G Ethernet is composed of four 10G Ethernet links. However, according to John d' Ambrosia, chairman of the Ethernet Alliance, many data centers have used servers of more than 10 Gb. This is why multiple chip manufacturers have provided 25 Gbit/s serial/destring transceiver. This not only facilitates the convergence of 25G, 50G, and G Ethernet bandwidth, but also reduces costs due to batch processing.

50 GB

Although there is still some time before the implementation of the 50G Ethernet IEEE Standard (around 2018-2020), many industry alliances expect that products will begin to emerge in 2016. Similar to 25g technology, 50G Ethernet technology will be the next solution to connect servers and data centers at high speed. According to data from the analysis company Dell Oro, in the next few years, servers and high-performance flash storage systems will require a speed of over 25 GB.

To facilitate faster delivery of these accelerated Ethernet technologies, the 25G/50G Ethernet alliance has free of patent fees for 25G and 50G Ethernet specifications and is open to all data center ecosystem vendors.

Reusing the 25g component of the existing G network can reduce the implementation cost of 50g. For example, if the cost structure of 25g cabling is the same as that of 10g cabling, the performance is 2.5 times. Similarly, the cost of 50 GB is half the cost of 40 GB, but the performance can be improved by 25%.

100 GB

For long-distance carrier networks ranging from several hundred kilometers to tens of thousands of kilometers, the deployment of G Ethernet will continue to grow.

However, according to the information provided by a new industry alliance, the G architecture will become another excellent market alternative. The 100G CLR4 alliance, jointly led by Intel and Arista Network, believes that 100G is ideal for connecting large "ultra-large scale" Data Centers between and 2 km.

Other companies are also looking for alternative GB implementation solutions for data centers. Avago Technologies has joined the CWDM4 MSA Industry Alliance, which defines a common specification for low-cost GB optical interfaces within two kilometers of data center applications. With the transformation of network infrastructure to GB Data Rate, data centers will require long-distance, high-density, GB embedded optical connections. MSA uses coarse Wavelength Division Multiplexing (CWDM) technology to provide four 25G Single-Mode Optical Fiber (SMF) link channels. Similarly, the OpenOptics MSA organization launched by Ranovus and Mellanox Technologies will focus on developing a data center that supports 2 km and GB.

In the past, the speed improvement promoted the development of most network components. Today, to process massive data streams through the cloud, companies need to seek a balance between speed improvement and reuse technology to find a cost-effective solution.