Driven by the bandwidth requirements of private and public cloud data centers and communication service providers, 100G experienced significant uptake, making network upgrade a necessary and inevitable trend. Until now, the majority of server vendors have already offered 25 Gigabit Ethernet NICs as the standard I/O option in their servers, driving the latest Ethernet speed transitions from 10G to 25G, 100G and beyond. This post will highlight the evolutionary path of 25G, then give a comparison of 10G-40G-100G and 10G-25G-100G network migration paths to help users make better choices for network upgrade.
While 100G and beyond is the next major milestone in hyperscale data center speeds, 40G may yet still have longevity in the industry. Traditional enterprises with tier 2 and tier 3 data centers that have lower density requirements and smaller budgets can reliably leverage the 40G architecture with 10G service speeds as the current building block.
Official development of 40G Gigabit Ethernet began in 2008, and is approved in 2010. It enables the transfer of Ethernet frames at 40 Gigabit per second and runs on Quad Small Form Factor Pluggable (QSFP+) cabling. In data centers, 40G switches are used as the aggregation/leaf switches, while they can act as core/spine switches for small and medium enterprises.
Though 40G has been active on the market for a time, its development faces difficulties. Ethernet has seen a huge increase in data speeds in recent years, some network operators are hesitating about deploying 100G systems. Different from 40G, 100G has done a better job in standard follow-up, technical solution unification and industrial chain development.
In 10G-40G-100G migration path, 10G equipment and cabling systems invested in the past will be utilized. Such a migration path can cost a lot, which can be analysed from the point of transceiver modules being used. 40G QSFP+ module is composed of 4 parallel 10G channels and 100G QSFP+ modules support a parallel channel rate of 10×10G. Therefore, in this type of 100G migration, the single-channel rate is based on 10G, while the migration path from 10G to 40G to 100G requires more fibers, resulting in more complicated cabling and higher cost. On the other hand, when QSFP modules are used in longer distance, such as QSFP+ LR4 optical module, they integrate 4 channels of 10G into one optical fiber link through wavelength division multiplexing (WDM). And this is part of the reason why the cost of QSFP optical modules remains high, especially for 40G modules with a longer transmission distance, such as QSFP+ ER4, which also increases cost for 100G network upgrade.
Data centers are expanding at an unprecedented pace, driving the need for higher bandwidth between the server and switches. To cater to this trend, networking and the Ethernet industry are moving from 10G to 25G which offers significant density, cost and power benefits for server to top-of-rack connections.
Ever since the 25GbE was initially proposed in 2014, though younger than 40G, it has been adopted and pushed by companies like Google, Microsoft, Arista, and Mellanox, etc. for top-of-rack server networking. These 25G network switches are often used as the ToR or leaf switches to connect the servers and terminal equipment. 25G Ethernet is an incremental update from 10G Ethernet. The high performance 25G SFP28 uses single-lane 25G SerDes technology similar in operation to 10G SFP+. These benefits allow existing 25G switch architectures to support link speeds faster than 10G with no increase in cable/trace interconnect, while keeping pace with the faster and richer growth trajectory of networking bandwidth, which helps reduce the power consumption and cost. Figure 1 highlights some of the key milestones of the 25G network from the year of 2014 to 2018.
The single-channel 25G speed is 2.5 times the transmission efficiency of 10G, while in a 100G network, only 4 channels 25G modules are needed to achieve the 4×25Gb/s mode. This will greatly save fiber links, but this migration path will eliminate the original 10G equipment in the system. Even so, 25G servers and 100G switches have become ubiquitous in hyperscale data centers, gradually replacing previous 10G servers and 40G switches. This speed migration has boosted overall system throughput by 2.5x with small incremental costs. As the Ethernet industry continues to innovate and lay a path to higher networking speeds like 200G and 400G, the 25G-100G upgrade has been developed as an inevitable roadmap for future data centers.
Before the 25GbE specification was released, the 10G-40G upgrade path was predominantly adopted as an option for enterprises, service providers and data centers to scale beyond 10GbE. However, with the emergence of 25GbE, the 25G-100G upgrade path has gained more momentum. What is the difference between 25G and 40G upgrading paths? Should I upgrade 10G to 25G?
It can be easily found that there are some obvious advantages in the process of 10G-25G-100G migration compared to 10G-40G-100G path:
25G-100G migration path can offer both CapEx (capital expenditures) and OpEx (operational expenditures) savings through backward compatibility, for investment protection and seamless migrations with consistent rack-design and reuse of the existing cabling infrastructure, avoiding costly and complex changes.
Single-lane 25G SerDes technology utilized in 100G Ethernet to 25G connectivity is similar to that in 40G to 10G connectivity, but the performance is increased by 2.5 times, thus reducing the power and cost per gigabit significantly. This power savings will in turn result in lower cooling requirements and operational expenditure for data center operators. By reusing the existing cabling infrastructure, the 25G Ethernet enables seamless network migration to avoid costly and complex changes.
25G Ethernet provides higher port and system density than a comparable 40G solution. The 25G-100G (4x25G lanes) networking migration path provides a lower cost per unit of bandwidth by fully utilizing switch port capabilities when compared to 40G-100G upgrade path.Therefore, upgrading directly from 40G to the faster 100G Ethernet is cost prohibitive compared to the 25G Ethernet connectivity.
The most important condition when planning to upgrade your network is to consider your business scale and future expansion. For some leading and giant companies, upgrading from 25G to 100G is the optimal way, as 25G has its obvious advantages which can satisfy the growing needs of the network and bring long-term benefits. While for some small and medium enterprises deployed with 40G network, there is no need to eagerly pursue a 10G-25G-100G migration path if the current network can fit your needs enough because running additional connections by buying more switches and cabling can be a big cost.
The demand for higher Ethernet speed and performance never stops for future data centers. By offering more bandwidth and higher port density with reduced power consumption and cost, the 10G-25G-100G Ethernet network upgrade path overturns the traditional 10G-40G-100G connectivity to improve data center efficiency, thus laying the foundation for further upgrade to 200G and 400G.