Written by Belden

 

Data center bandwidth continues to climb—and it’s not coming back down any time soon.

 

New technology and applications call for faster speeds. From where we stand today, the industry is moving quickly toward 400G, 800G and even 1T connections (and higher). According to Henkel’s 2024 Data Center Pulse Report, nearly 70% of data center and telecom professionals around the globe have adopted some form of 800G technology (with varying levels of success), and 33% believe the industry will fully transition to widespread use of 1.6T in just one to two years.

 

What does this mean for data centers in the future? For starters, faster speeds equal more fiber. In Getting Granular on Connectivity, a recent webinar led by Data Center Dynamics, Belden Vice President of R&D and Strategy Andrew Oliviero shared his thoughts on data center trends. During the discussion, he explains and explores the connection between data center bandwidth and the need for more fiber.

 

We recapped some of the highlights below, but you can watch the full discussion any time.

Making Data Center Predictions

To support higher data rates, data center trends will involve many upcoming changes. During the Data Center Dynamics webinar, we discussed some of our predictions:

• As lane rate per wavelength starts to plateau, duplex links are being replaced with parallel links that can offer more fiber-pairs per link.
• Instead of Top of Row (ToR) layouts, we’ll see more Middle of Row (MoR) and End of Row (EoR) switches with optical switch-to-server interconnects, especially in cloud and hyperscale environments.
• Historically, data center switch investments have been focused on front-end networks. With the increasing use of AI, however, data center workloads will require new back-end infrastructure driven by high-speed Ethernet or InfiniBand transmission protocol.

 

Multimode and Singlemode Fiber Both Matter

According to the Ethernet Alliance’s Ethernet Roadmap, singlemode deployment hovers between 50% and 60%, and it rises every year.

 

As a result, some industry professionals have been anticipating a slowdown in multimode fiber installation. For now, however, this data center trend isn’t taking hold. Multimode fiber is keeping pace with singlemode—and the Ethernet Roadmap indicates that it will continue to do so for quite some time. For example, many of today’s hyperscale data centers still use multimode fiber links for some of their applications.

 

Although multimode is still relevant, it’s critical that data centers begin to prepare now for the increase in singlemode fiber that will be necessary to accommodate more data transmission and faster speeds.

 

To support these connections, data centers will also need the right components in place. Future networking technologies will continue to drive the importance of high-density fiber optic cable and connectivity, such as:

• High-density optical distribution frames that support optimal space utilization, scale as business grows and changes, protect signal integrity and support multiple fiber configurations.
• Multifiber connectors (MPOs) and cassettes that can support the rapid deployment of high-density infrastructure.
• Ribbon fiber, which saves space and installation time in data centers.
• Breakout assemblies that can break out to multiple ports to support varying speeds.
• VSFF (very small form factor) connectors that feature a compact, space-saving design to improve data center density.

 

Changes in Network Architectures

With more data and faster speeds also come different network architectures.

 

Especially in enterprise networks, devices have historically been connected to aggregation switches that then connected to a layer 3 core switch and router.

 

This traditional three-tier hierarchical architecture accommodates traffic between servers that are connected to the same access switch; however, traffic between different access switches transmits through higher-level switch tiers in a north-south pattern. This creates speed and latency differences, which can lead to performance problems in data centers.

 

Today, the industry is migrating to a much simpler, two-tier architecture, where aggregation switches connect directly to spine switches. Called leaf-spine architecture, this approach eliminates a layer of hardware, improves latency and supports better port utilization.

 

Every leaf switch connects to every other leaf and spine switch. To evenly distribute traffic among top-tier switches, the transmission path is chosen randomly. If a switch fails, then performance degradation is minimal (but there are also more fiber connections that must be managed).

 

To connect leaf and spine switches, cross-connect architecture is a popular approach. It uses patch panels that mirror the leaf-switch ports and are connected by permanent cabling instead of by patch cords. This allows data centers to easily change which leaf switch port connects to which spine switch port.

 

Optimizing Data Center Investments

As data center bandwidth gathers speeds, Belden will continue to work with its customers to find systems and innovations that support holistic and client-specific solutions to enable scalability, customizability, high density and modularity.

 

We excel at not only helping data centers prepare for higher speeds and new media types but also finding ways to optimize investments in pre-existing infrastructure to reduce future expenses.

 

Find the original article here