English
Deutsch
Espaniol
1. Executive Summary of the development of the 25G Standard
The ever-increasing demand of diverse multimedia applications and services as well as the requirement for bandwidth expansions and faster data rates becomes a challenge for every data center everywhere. Emerging technologies like cloud computing also acted as catalyst in driving the industry to develop a new approach to adopt in these fast-charging trends.
Reliance on networking permeates every aspect of our world, and data center bandwidth requirements are expanding at double-digit rates-along with an equally urgent push to contain costs. New technologies necessitate that data centers remain flexible and scalable enough to adapt to changing requirements. The rise of cloud providers changed the data center Ethernet landscape, creating a viable market for high-speed, reasonably-priced connectivity.
Leading cloud and telco providers are clamoring for even more network performance in order to meet the needs of their web-scale data centers and cloud-based services, without compromising the cost-to-performance ratio. To help address network performance needs, leading manufacturers have joined forces to define and drive the 25 Gigabit Ethernet (25GbE) technology.
To suffice the increasing demands of collaborative multimedia services and applications, to answer to the fast-changing traffic patterns, and to improve accommodations of users’ bandwidth requirement for communication. This is one of the reasons why an industry consortium was formed to create a new Ethernet connectivity standard in data centers. The consortium’s goal is to enable the transmission of Ethernet frames at 25 or 50Gb per second (Gbps) and to promote the standardization and improvement of the interfaces for applicable products. Last July 2014, the IEEE agreed to support the development of this 25GbE standards for servers and switching due to the increasing demand for a much faster network performance while maintaining Ethernet economics. This standard was called 25 Gigabit Ethernet or 25Base-T, developed by the IEEE 802-3 task force P802.3by. This standard was derived from 100GbE, since its operation works with four 25Gbps that are running on four fibers in each direction. The IEEE 802.3by 25GbE standard is technically complete and ratified on June of 2016.
2. The Push for higher Bandwidth for Media and Transceiver Modules
The requirements in the market for Ethernet are constantly changing for different applications and the coherent need of speed, further distance and lower costs. Various speeds are also needed for specific applications like Wireless Access Points that use 2.5GbE and 5GbE; servers which need up to 25GbE and lastly core networks which operate with up to 400GbE.
Due to this exponentially growing, global bandwidth requirements, Ethernet speeds are also in constant development at a high rate to stay ahead of these demands. However, innovation is occurring within varied application spaces at lower speeds. Most of todays’ servers are still using GbE, and some users do not care for a foreseeable future about higher speeds like Terabit Ethernet (TbE) and 400GbE.
There is diversification of efforts to coupe up these requirements, but a common goal will undermine the diversities, which is a global requirement for a market-driven standard, fostering innovation and enabling multi-vendor interoperability across whatever application area the world’s growing cast of Ethernet users seeks to enable.
Ethernet has always been and will always be about connectivity and how far this technology can go, but the Ethernet community embraces that the need for speed is relative to the given application.
In years, since 1995 until 2010, the Ethernet’s evolution was somewhat slow in pace and innovation was mostly simple. Ethernet speed increased linearly - approximately an order to an extent every few years like 10 Mbit/s to Fast-Ethernet 100 Mbit/s and from 1G to 10G. Consequently, around 2010, the first 100G Ethernet version 100GBASE-SR10 was introduced. Below are some of the trends for this development.
25GBASE-SR: Over the development of 100GBASE-SR4 in 2015, it became deem compulsory to develop the intermediate speed 25GBASE-SR to comply on standards as IEEE 802.3 by 2016. This transceiver utilizes the popular SFP+ form factor with LC-Duplex connector interface but with 25G speed. Meanwhile known as SFP28 and known to have collective four 25G server ports at one 100G switch port.
50GBASE-SR: Designed to combine four 50G server ports to a 200G switch port. Available in the market since 2018. The transceiver has the SFP+ form factor with LC-Duplex connector interface, but in this case it runs with 50G speed and is called SFP56.
100GBASE-SR2: The objective of this standardization project is to aggregate two 50G server ports at one 100G switch port. This Ethernet version will be available along with the 50GBASE-SR. The transceiver will have the QSFP form factor with LC Duplex connector interface.
200GBASE-SR4: It will become available to the market together with 50GBASE-SR4 and 100GBASE-SR2. The configuration is expected to conform with the earlier series of Ethernet SR4 devices.
Based on the released road map of the Ethernet Alliance in 2015, which outlines the response of Ethernet to the ever-increasing desire for higher bandwidth in data centers, Ethernet speed is shown to have unprecedented level of activity in the low end of the market, this roadmap also shows Ethernet speed in the future.
After the ratification of IEEE 802.3by 25GbE in June 2016, we can now see that a new Ethernet speed becoming the new common standard. 25GbE which is designed to replace 10 GbE because of its cost effective and power efficiency to various application like ToR switching for cloud providers. This new trend will also help the demand for the huge data volume as well as the speed needed for the internet of Things. Furthermore, this development focuses primarily to the fiber-based SFP28 and QFSP28 market for the purpose of backbone or longer-haul connections.
3. The Advantages of a QSFP28 Transceiver Solution
The QSFP28 transceiver is projected as the prospective interface by the Ethernet Alliance. This transceiver enables network bandwidth to be cost effective and resource saving. It is designed for 100GbE speeds using the 4x25GbE wiring specification, hence the “Q” which stands for quad. The QSFP28 form factor retains the usual density of 48 ports in a 1U tall switch which is very advantageous to existing systems without the need to migrate the network to a new standard. Moreover, QSFP28 increases the density but minimizes the power consumption. QSFP is slowly becoming the universal form factor for various reasons.
First, it increases front panel density by maintaining the form factor and the maximum number of ports but increases the lane speed from 10Gbps to 25Gbps. Secondly, it supports both cables and transceivers. Using QSFP28, a one-rack unit switch can accommodate up to 36 QSFP ports. Lastly, QFSP28 can use either VCSELs for short distances or Silicon photonics for longer distances to support data center to reach for more than 2 kilometers of interconnection.
With all these benefits coming to a single form factor, the next versions of high-bandwidth switches, routers and adapters will feature QSFP28 ports to ensure 100GbE data center interconnection.
4. 25GbE SFP28, today’s new Transceiver Standard
25GbE is already an emerging standard for Ethernet connectivity that will be beneficial to cloud and enterprise data center environments. Due to the Consortium that was established in June 2014 an Ethernet Task Force was created to encourage 25GbE technology and successively, to further develop the standard - the IEEE P802.3by. Moreover, the IEEE P802.3bq 40GBASE-T Task Force adopted objectives to likewise improve BASE-T support for 25GbE.
There are various market drivers why 25GbE standard emerged. The main reason is, it provides a server connection speed faster than 10GbE that is optimized for cost, throughput and efficiency. Moreover, it maximizes efficiency of server connections to access switches in data centers. Lastly, this leverages four 25Gbps lanes (IEEE 802.3bj) running on four-fiber or copper pairs individually transmitting at 25Gbps. This sums up to a backplane of 100GbE per form factor. Every lane needs a Serializer / Deserializer (SerDes) chipset. This twisted-pair concept originates from the 40GbE standard development. The below table shows a synopsis of fundamental upcoming IEEE standard interfaces that specify 25GbE.
IEEE 802.3 Standard Interfaces that specify 25GbE
|
Name |
Error Correction |
|
|
MMF Optics |
25GBASE-SR |
RS-FEC |
|
Direct Attach Copper |
25GBASE-CR |
BASE-R FEC or RS-FEC |
|
Direct Attach Copper |
25GBASE-CR-S |
BASE-R FEC or disabled |
|
Electrical Backplane |
25GBASE-KR |
BASE-R FEC or RS-FEC |
|
Electrical Backplane |
25GBASE-KR-S |
BASE-R FEC or disabled |
|
Twisted Pair |
25GBASE-T |
N/A |
There are various fundamental benefits of 25GbE, it allows network bandwidth to be cost-efficiently scaled in support of succeeding generation server and storage solutions existing in cloud and web-scale data center settings. Below are its most noticeable benefits.
1. Reduced CAPEX
- Lower Cost compared to 40GbE
- Fewer ToR switches and cables
2. Maximum switch Input / Output performance and fabric capability
- Four times the switch port density versus 40GbE (one-lane versus four-lane)
- Higher performance than the existing 10GbE
- A single lane per physical port maximizes the number of connected servers or uplinks per switch
3. Fast maturation by leveraging the current IEEE 100GbE standard
4. Reduced OPEX
- Lesser power, cooling and smaller footprint requirements
Some members of the 25 Gigabit Ethernet Consortium who are prominent suppliers of Ethernet switching solutions are now offering 25GbE-capable Ethernet switch platforms. Most Ethernet switching solutions including 10GbE, 25GbE to 100GbE support multiple Ethernet rates, which means consumers have absolute cable choice for network connectivity. Some of the notable vendors who use multispeed 10/25/40/50/100GbE switch platforms that offer these emerging standards are Cisco, Arista, Broadcom and Mellanox.
In addition, Network Interface Cards (NIC) for the 25GbE standard are also rapidly released. As a result, cables are also adopting this emerging standard. It is essential that 25GbE and 50GbE channels compel on all the “channel characteristics” described in IEEE standards under 802.3bj, section “Physical Medium Dependent (PMD) sublayer and baseband medium, type 100BASE-CR4,” section 92.9., this is according to the 25Gb Ethernet Consortium. A number of converter and cable combinations can meet the features.
Specifics of 25GbE and 50GbE PMDs are both low-cost and with twin-axial copper cable availability. 25Gbps operation however, requires only two twin-axial cable pairs whereas 50Gbps needs only four-twin-axial cable pairs.
ToR switches usually connect to servers via a link based on copper twin-axial cables as well as the intra-rack connections between switches and/or routers.
Moreover, cables that connect to higher speeds and “fan out” (to multiple lower speed links) are able to connect to 10/25/40/50Gbps speed which is now possible by the use of multi-mode or Single-mode fibers, copper cables, matching reach-range to the specific application need.
|
Type of Technology |
Media |
Standards |
Power |
Reach Range |
|
100GBASE-SR4 |
Multi-mode |
100GbE/25GbE |
3.5 w |
100 meters |
|
100GBASE-LR4 |
Single-mode |
100GbE |
4.5w |
2km or 10km |
|
PSM4 |
Parallel Single-mode |
100GbE/25GbE |
3.5w |
500 meters |
|
CWDM4 |
CWDM single-mode |
100GbE |
3.5w |
2 meters to 2km |
|
100GBASE-CR4 |
Passive Copper |
100GbE/25GbE |
3.5w |
5 meters |
A Technical Overview of the 25GbE SFP28
CLOCK RATE
Inside Ethernet NICs or switches, a serial component called SerDes connects all of the high-speed components which take data to transmit and then serialize it. Then, a Deserializer on the recipient side reconstructs the serial stream of bits into data for the final receiver. Over the years, the SerDes technology advanced to the newest 25Ghz rate.
The mechanisms that consist of 10GbE switches run at 12.5Ghz SerDes with a clock-rate of 10.3125GHz. Nevertheless, the present 40GbE NICs and switches utilize four parallel SerDes links having a clock rate of 10.3125GHz each. In the contrary, the components encompassing the 25GbE NICs and switches use a single lane SerDes with a clock rate of 25.78125GHz. Below is a table that shows the clock rates, lanes and performance 25GbE SFP28.
|
Clock Rate |
Lanes |
Data Rate |
|
|
1 GbE |
1.25 GHz |
1 |
1 Gbps |
|
10 GbE |
10.31 GHz |
1 |
10 Gbps |
|
25 GbE |
25.78 GHz |
1 |
25 Gbps |
|
40 GbE |
10.31 GHz |
4 |
40 Gbps |
|
100 GbE |
25.78 GHz |
4 |
100 Gbps |
Number of transmission connections with SerDes
The actual Ethernet ports in a usual 10GbE/40GbE Ethernet switch are SerDes connections coming from the switching chip pins, which were used to connect directly to the SFP+ and QSFP optics cages or other Ethernet or fabric chips (for blade servers). Communication amid an SFP+/QSFP in the front of the switch and the switching chip runs on top of one of these SerDes connections. With this, the term “Lanes” is used to call the number of SerDes connections required to drive a switch port.
Currently switches use components that are all run by SerDes with a clock rate of about 10Ghz, delivering a 10Gb transfer rate among every component, permitting for the encoding overhead. SerDes technology improved in the past several years that it reached 15Ghz SerDes, it turned out to be economically doable and all of the different physics linked challenges in signal integrity found dependable solutions.
Four parallel SerDes links between the Ethernet chip and the QSFP pluggable module is the composition structure of the so-called 40GbE interface. It remains indispensable to have four parallel 10Gb streams in extending QSFP onto fiber to transport this to the receiving QSFP (i.e. parallel optics). Short reach QSFP interfaces utilize four pairs for the transmission. Long-reach QSFP interfaces use an internal Coarse Wave Division Multiplexing (CWDM) to transport the four 10Gb streams over a single pair of fiber. The requisite of four lanes substantially decreases switch port density per switching chip and escalates the cost of cabling and optics.
The 25GbE standard leverages the availability of a 25Ghz SerDes and requires only a single SerDes lane, while delivering 2.5 times more throughput compared to 40GbE solutions and significant CAPEX savings compared to 40GbE solutions.
Moreover, existing several blade server chassis solutions have limits of only two SerDes lanes for their LAN on Motherboard (LOM) networking ports, hence, they can’t implement a four-lane 40Gbps interface.
IEEE error correction code for 25 Gigabit Ethernet
Starting with 10GbE and 10Gb Fibre Channel Inter-Switch Links (ISLs), the “64b/66b” encoding scheme is used to develop data transfer efficiency.
The 64b/66b encoding outcomes a 3% overhead (66-64) /66 on the raw bit rate. To pay off, Clause 74 (Fire code) FEC was presented to deliver extra error protection.
25GbE specification both supported Clause 74 FEC and Clause 91. Auto-negotiation can determine whether Clause 74 FEC, Clause 91 FEC, or no FEC is employed on the link.
Auto-Negotiation for 25 Gigabit Ethernet
Specifics of auto-negotiation capabilities aren’t fully established or implemented. The 25GbE and 50GbE solutions are nevertheless backward and forward compatible with 10GbE, 40GbE, 100GbE and 400GbE products since they use the same IEEE 802.3 frame format. However, switch port capability to automatically select a different speed is still under development.
Different Form Factors for 100G and 25G
Shown here below, the 25GbE physical interface description supports various form factors:
|
Lanes and Speed |
|
|
QSFP28 |
4 x 25Gbps |
|
SFP28 |
1 x 25 Gbps |
Currently some available Switches don’t support direct 25GbE connections by means of a SFP28 Direct Attach Copper (DAC) cable. The usage of a breakout cable that permits 4x 25GbE ports to connect to a 100GbE QFSFP28 switch port is one of the suggested solutions. Lengths of Direct Attach Copper (DAC) cable are limited to five meters for 25GbE. To support longer lengths, Active Optic Cable (AOC) solutions can be utilized as well.
PCI Express (PCIe) Interfaces
The PCIe 3.0 interface is present-everywhere transversely in shipping server platforms. Due to cost, the preference in cloud and web-scale server deployments is headed for single-port Ethernet connectivity. These volume servers usually have PCIe 3.0 x4 slot(s).
The table below shows why 25GbE is an easier move to upgrade from 10GbE lanes for it entails half the number of PCIe lanes and fits into the existing model against 40GbE lanes, resulting to better PCIe bandwidth usage and lower power impact.
PCIe 3.0 Lanes required for Ethernet Generations
|
Single Port |
Dual Port |
|
|
10GbE |
2 |
4 |
|
25GbE |
4 |
8 |
|
40GbE |
8 |
16 |
|
100GbE |
16 |
32 |
Conclusion: 25GbE replaces 40GbE
The consortium held in July 2014 was a call for interest of members who unanimously agreed to support the development of a new standard for servers and switching. They created the 25Gbps Ethernet Task and developed the 25Gbps standard. There is a fast progress of this standard due to the high leverage of the existing standard for 100GbE as the base standard. The new 25GbE standards maximizes server efficiency to access switch interconnects and provides an opportunity for optimum cost/performance benefits. It will provide up to 2.5 times faster performance than existing 10GbE connections while maximizing the Ethernet controller bandwidths/pin and switch fabric capability. More than 50% of the rack interconnect cost per unit of bandwidth can be saved and significantly improve an operator’s bottom line. Furthermore, it increases network scale and accommodate higher server density within a rack than what is currently achievable with 40GbE ToR links. In short order, deploying 25GbE solutions, something that solutions built on the more complex 40GbE standard can never achieve.
