Optimize Data Center Interconnect With Lambda Splitting to Support AI Workloads

The Evolution of Fiber Optic Efficiency

Imagine a bustling highway system where traffic must flow as efficiently as possible. To prevent congestion, vehicles are distributed across multiple lanes, ensuring smooth movement and maximum throughput. This challenge mirrors modern data centers, where skyrocketing AI-driven traffic demands smarter ways to utilize available fiber capacity. One such approach is Lambda Splitting—a transformative method that enables the efficient transport of high-speed Ethernet clients across multiple wavelengths. By intelligently distributing traffic, this technique plays a crucial role in maximizing fiber efficiency and adapting to the ever-growing needs of optical transport infrastructure.

In this blog post, we will delve into how Lambda Splitting optimizes Data Center Interconnect (DCI) by increasing the bandwidth that can be transmitted across the coherent optics and optical fiber by up to 50%. We'll discuss the challenges posed by fixed router port rates, the benefits of distributing high-speed Ethernet clients over multiple wavelengths and how solutions like META-DX2+ facilitate this process. By the end, you'll understand how Lambda Splitting enhances fiber reach, reduces costs and maximizes fiber efficiency in modern networks.

Background and Industry Trends

The networking industry is undergoing significant transformation. The rapid adoption of cloud computing, AI-driven workloads and growth of hyperscale data centers has led to an increased demand for high-speed Ethernet connections, notably 400GE and 800GE. While router and switch ports are optimized to operate at fixed rates, coherent optics between data centers support variable data rates based on the distance and other properties. Therefore, transport networks require the right solutions to avoid bandwidth wastage and bottlenecks.

Recently, network operators have leveraged both pluggable optics and transponders to address diverse connectivity needs. Short-reach DCI connections often utilize optics plugged directly into switches or routers, while transponders continue to play a key role in extending reach and providing demarcation across various applications, including metro and long-haul networks. As network demands evolve, optimizing fiber utilization remains critical. Advances in pluggable coherent optics have expanded the range of deployment scenarios, offering an additional option alongside traditional transponder-based solutions to balance cost, power efficiency and operational flexibility.

Pluggable Coherent Optics

The 400ZR Implementation Agreement (IA), defined by the Optical Internetworking Forum (OIF), is designed primarily for point-to-point DCI applications. It supports distances up to approximately 120 km, depending on fiber conditions. It employs pluggable coherent optics in QSFP-DD and OSFP form factors, enabling high-density deployments due to its small size. The 400ZR+ Multi-Source Agreement (MSA) extends the capabilities of 400ZR by supporting longer distances and multiple client types. It supports higher transmit power, enhanced FEC and more flexible modulation schemes, making it suitable for deployment in metro and regional networks.

The OIF 800ZR IA doubles the data rate per wavelength compared to 400ZR. 800ZR+ supports line rates of 400G, 600G and 800G. These technologies follow the same principles as 400ZR and 400ZR+ but incorporate improvements in DSP rate and modulation techniques.

Generally higher data rates translate to lower reach, and that is primarily based on changing the modulation format. Higher-order modulation formats that carry more bits per symbol are more sensitive to noise and fiber impairments, as a result reducing reach. 1.2T optical modules are well suited for high-capacity DCI applications, while for longer distances the 1.2T module can be configured to operate at a lower rate, such as 1.0T, 800G, 600G or 400G to extend the reach.

Running optical modules at 1.0T or 600G rates does not optimally align with datacenters that interconnect their switches and routers with 400GE interfaces. An optical module running at 600G line rate can only transport one 400GE client, leaving 200G of unused line capacity. However, an additional 400GE client could be transported if it was “split” between two 600G wavelengths. As network operators aim to optimize existing infrastructure cost-effectively, Lambda Splitting has emerged as a compelling solution.

Lambda Splitting for Efficient Fiber Utilization

Lambda Splitting enables large Ethernet clients to be transported efficiently over multiple wavelengths, ensuring maximum fiber utilization without backpressure or subrating issues at the router or switch. Lambda Splitting has FlexE Bonding-like functionality that enables bonding across multiple wavelengths. It is similar to previous generations of FlexE technology, without requiring FlexE support on the router or switch.

The image below demonstrates the tradeoff between data rate and signal reach. The higher the data rate, the shorter the transmission distance. In the example shown, two 1.2T coherent modules are used to transmit the signal, and at each of the rates shown, the maximum number of the 400GE clients that can be deployed are listed in blue.

For a link with the distance and properties to support a 1T wavelength, standard 400GE client deployments would typically utilize only 80% of its capacity, supporting 2×400GE connections. By using lambda splitting, an additional 400GE client can be divided into two 200G segments, with each 200G portion allocated to separate underutilized links. This allows both links to reach their full 1T capacity, effectively increasing total bandwidth utilization by 25%.

In a system with two 600G wavelengths, a maximum of two 400GE clients can be transmitted. However, by employing Lambda Splitting, the same two 600G wavelengths can support three 400GE clients, increasing traffic throughput by 50%, while maintaining alignment and integrity of the client over the link. By using 800ZR+ pluggable optics at a 600G line rate, this method enhances reach while minimizing the number of required optics.

Another key use case of Lambda Splitting is transporting an 800GbE connection over two 400G wavelengths. The following configuration leverages existing 400ZR or 400ZR+ optics, enabling network operators to upgrade routers to 800GE incrementally without overhauling transport optics.

Benefits of Lambda Splitting

Lambda Splitting is an efficient method to send a high-bandwidth client through multiple lower-rate, IEEE-compliant Ethernet PHYs over coherent links. The splitting and recovery process ensures reliable transport, deskew and reassembly of data. This approach maintains timing transparency, meaning that there is no impact on synchronization and network timing. Additionally, MAC layer alarm transparency ensures the clear transmission of alarms, status indicators and consequential actions to maintain end-to-end quality of service. For example, remote fault alarms allow network operators to detect link failures early, preventing unnecessary data loss and reducing downtime. By providing real-time fault detection, Lambda Splitting enhances network reliability and ensures uninterrupted data transmission.

By leveraging Lambda Splitting in the META-DX2+ device, network operators can optimize data transport capacity by adapting Ethernet clients to available wavelength capacity. This maximizes fiber utilization and reduces the need for optical regeneration.

Conclusion

Lambda Splitting is revolutionizing DCI strategies, offering a cost-effective and scalable approach to high-speed networking. When META-DX2+ is used with coherent optics in optical transport equipment, network operators can maximize fiber utilization and enhance reach in their networks, therefore significantly reducing the need for additional optics and regeneration sites, streamlining operations and lowering costs.