All Articles

25G networking has quietly become the building block of enterprise and data center network upgrades. The current wave of 100G upgrades are built on 25G lanes delivering cost-effective 100G networks. 40G networks will remain a staple in the enterprise and the data center for the short term, but 25Gs role in building future networks is bright.

25G upgrades in the enterprise and data center offer advantages over 40G. 25G, SFP28 transceivers offer the benefits of the SFP and SFP+ form factors (low power consumption, common footprint, and density) and use existing 10G cabling. Sporting a single 25G lane, the SFP28 also rolls nicely into future network upgrade standards. 100G QSFP28 and 200G QSFP28-DD each use multiple 25G lanes in their architecture.

In contrast to 40G, 25G is being introduced directly to the edge NICs and switches, rather than a line side upgrade. Unfortunately, simply installing an SFP28 transceiver into an SFP switch port does not simply upgrade a port to 25G. Edge devices must be upgraded to support 25G. Upgrading the sheer volume of edge devices in a network will require both budget and time. Advances in transceiver technologies are now available to allow enterprises and data centers to upgrade only portions of their network to 25G while deferring upgrades on other portions into the future.

The new SFP28 10G/25G transceiver is dual rate, capable of connecting at 25G or 10G. When installed in a distribution or aggregation switch, the SFP28 10G/25G transceiver can connect over existing network cabling with either a 10G SFP+ or 25G SFP28 transceiver installed in edge devices.

This approach allows data centers and enterprises to align 25G upgrades of edge devices and NICs with time and budget constraints. SFP28 10G/25G transceivers are available in SR (OM3/OM4 multi-mode fiber) and LR (single-mode) variants.

Demonstrating the versatility of 25G, breakout applications separate the transmit and receive pairs from a QSFP transceiver to four SFP-type transceivers. QSFP breakout applications are prevalent in top-of-rack, end-of-row and spine/leaf environments. In 100G networks, aggregating four 25G network devices into one QSFP28 port offers power, port and reduced latency advantages over one-to-one port approaches. 100G to 25G breakout applications are supported by multiple solutions.

Transceiver based solutions utilize existing fiber cabling to avoid costly plant upgrades. The QSFP28 SR4 transceiver accepts an OM3 or OM4 MPO-type connector with various breakout cables or cassettes that separate the physical lanes into four transmit and receive pairs to interface with SFP28 SR transceivers. The QSFP28 PSM4 transceiver takes a similar breakout approach but requires an angle polished single-mode MPO-type connector cable to interface with SFP28 LR transceivers. Transceiver breakout solutions are typically deployed in end-of-row and spine/leaf applications, leveraging existing 10G fiber cable infrastructure.

Direct Attached Cables (DACs) and Active Optical Cables (AOCs) are ideal solutions for new top-of-rack and end-of-row QSFP port breakouts. DAC and AOC breakout cables have a pre-terminated QSFP-type transceiver on one end with four SFP-type transceivers on the other. 100G DAC breakout cables are widely deployed in top-of-rack applications, connecting a top-of-rack QSFP28 switch port to four SFP28 ports. 100G AOC breakout cables are more common in both end-of-row and spine/leaf applications to overcome distance limitations of copper direct attached cables.

25G solutions offer enterprises and data centers an upgrade path aligned with future network technology while reducing the cost of network cabling upgrades.

Content written by Americas Product Manager, Ray Hagen

London, 19 November – The road to 200G and 400G is set to become clearer after LINX103 in London today, as Ambroise Thirion, Technology Solutions Specialist at ProLabs presents the company’s roadmap towards achieving 400G.

The past few years has seen ProLabs, industry leader in compatible optical transceivers, support its customers in data centres with bitrates of 25G, 50G and 100G. This trend of expansion to meet consumer demands is set to extend throughout 2019 and early 2020, with the recent development of form factor specific technologies to enable up to 400G.

This need for higher bitrate is being driven by the growing demand for bandwidth, the increased need for density and high-volume capabilities, all while minimising power consumption and power dissipation. With VOD and 5G, the number of connected devices worldwide set to increase markedly by 2020 and this demand is being felt industry wide.

“Upgrading and future-proofing networks can be daunting and costly, which is why it is so important organisations take the time to understand; why they are moving to 200G or 400G, how these bitrates can be achieved using the right technology and when the preferred solution can be expected to hit the market,” said Thirion.

“Some of the larger data centres are putting plans in place to upgrade to 200G as soon as its available, while another scenario we are seeing is that organisations are opting to wait for a more mature technology and looking for a direct and complete deployment with 400G. It is ultimately up to the organisation on which route to take and when to take it. This is where our roadmap to 400G, presented at LINX 103 today, can help,”

200G and 400G are achievable thanks to a number of technologies including; PAM4 modulation, GEARBOX for 400G and a set of new form factors such as QSFP-DD, OSFP and CFP8. As compared to the previous binary mechanism NRZ, PAM4 enabled transceivers offer customers double the transmission bitrate by increasing the levels of transmission of pulse-amplitude modulation.

For 400G specifically, the initial form factors will include eight lanes of 50G PAM4, while other versions will utilise the latest GEARBOX technology to facilitate four lanes of 100G PAM4. The GEARBOX component combines the power of numerous PAM4 modules to achieve the speed desired, in this case, 200G or 400G. Currently under development, ProLabs customers can expect solutions for up to 200G to be available from 2019 and 400G to be available from end of 2019.

Thirion added: “While there is no single best way to combine solutions for the ideal network, ProLabs have products available across a variety of form factors to suit specific applications. We have a proven track record in keeping our customers ahead of the curve, and the impending 200G/400G upgrade is no different. We are already supporting our customers in up to 100G networking and look forward to enabling 400G in the near future,”

What's the difference between 400GBASE-LR8 and 400GBASE-LR4?

Both QSFP-DD and OSFP transceivers can transmit at 400Gbps using 8 lanes of 50Gbps PAM4 or 4 lanes at 100Gbps PAM4. But what's the real difference behind 50Gbps PAM4 and 100Gbps PAM4? How do we achieve this transition?

Pre-requirements

Modulation

For all optical transceivers transmitting up to 100Gbps, a NRZ mechanism (binary non-return to zero) is used. In practical, we can transmit 1 bit per waveform (0 or I).

With PAM4 modulation (pulse amplitude modulation), each waveform can carry 2 bits (instead of 1 for NRZ). To achieve that, the waveform has 4 different levels, carrying 2 bits: 00, 01, 10 or 11.

Baudrate

The baudrate wording has not been used in optical transmission until now because there is an equivalence between baudrate and bitrate for NRZ transmission. A baud is the unit representing the symbol rate and is becoming significantly important for 400G transmission.

A transmission with NRZ mechanism will have the same baudrate and bitrate because one symbol can carry one bit. 25Gbps (gigabit per second) bitrate is equivalent to 25GBdps (gigabaud per second) baudrate.

Because PAM4 is carrying 2 bits per symbol, it is important to make a difference between baud rate and bitrate: 50Gbps PAM4 (50 gigabit per second with PAM4 modulation) will have a line transmission at 25GBdps (Gigabaud per second). It means that for 50Gbps PAM4, the symbol rate remains to 25 Gigabaud.

Parallel optic

A regular 100G QSFP28 doesn't transmit at 100Gbps but at 4x 25Gbps; it has, inside the module, 4 lasers and 4 receivers.

The actual 400G transceivers (QSFP-DD and OSFP) always have 8 lanes of 50Gbps PAM4 on the electrical side (400GAUI-8) and have 8 lasers of 50Gbps PAM4 or 4 lasers of 100Gbps PAM4. For the last version, an electrical conversion (Gearbox) from 8x 50Gbps PAM4 to 4x 100Gbps PAM4 is required.

The gearbox

The gearbox is a component converting a specific amount of data lines at a nominal bitrate to a lower or higher amount of data lines at another bitrate, following a ratio. The gearbox can do the same job in the opposite direction also called as "reverse gearbox".

The gearbox is not a new concept but is now getting a real importance for the 400G applications:

In the most recent designs, the gearbox is embedded into a DSP IC (Digital Signal Processor, Integrated Circuit) which can have additional functionalities, such as equalization, retiming, etc.

The gearbox presence can be easily identified following the optical standard: a QSFP56-DD 400GBASE-

LR8 doesn't have a gearbox while the 400GBASE-LR4 does.

Transmitting at 100Gbps PAM4

The signal integrity is becoming extremely challenging when transmitting at 100Gbps PAM4 per lane;

considerably reducing the maximum optical reach.

While 400GBASE-LR8 (8x 50Gbps PAM4) is now mature, the 400GBASE-LR4 (4x 100Gbps PAM4) is clearly on the edge and doesn't have any margin left. A strong FEC mechanism will be required to bring the communication at a correct BER for a regular 400G connection.

The above eye diagrams show a higher complexity for the 100Gbps PAM4 compared to a 50Gbps PAM4.

This higher complexity reduces the margin and brings higher requirements to avoid chromatic dispersion, jitter, etc. With regular technology, the maximum reach is lower with 4x100Gbps PAM4 compared to 8x50Gbps PAM4.

Conclusion

The right technology must be chosen following the needs and application. 4x100Gbps transmission is interesting as it only uses 4 lasers instead of 8. It lowers the power consumption, reduce the hardware complexity and relax the requirement on xWDM grid; however, it requires a gearbox and it's clearly bringing a challenge in term of reach.

If the gearbox cost is dropping down in the future, the 4x100Gbps seems more interesting for the datacentres, clearly for 2 different applications:

• A breakout solution from a single 4x100Gbps QSFP-DD (400GBASE-DR4, with MPO connector) to 4x QSFP28 100GBASE-FR (with LC connector); used in server-leaf architecture.
• A cost-efficient module for a max 2km reach over duplex singlemode fibres (400GBASE-FR4); used in leaf-spine architecture.

For link with a reach of maximum 10km, 400GBASE-LR8 is available and ratified under IEEE 802.3bs. The 400GBASE-LR4 is not ratified by the IEEE committee yet.