When the data center industry talks about the pressures that it has come under in recent years, through the advent of mainstream artificial intelligence (AI), along with the increasing pressure on power networks to meet the demands of ever-higher power servers, we usually look to expand, retrofit and densify. We look to liquid cooling and alternative energy sources. However, one vital aspect of the data center environment that is often overlooked is the all-important optical fiber cabling that criss-crosses data halls and provides the capacity to allow the traffic to travel both internally, and out into the great wide open beyond.

AFL is an expert in the field of optical fiber connectivity, and as it celebrates its 40th anniversary in 2024, the importance of understanding the vital role that interconnects play has never been more crucial. To this end, DCD spoke to Manja Thessin, market manager for market, strategy, and innovation at AFL to talk about the latest innovations in fiber, and why it’s more important than ever to ensure that data centers are maximizing capacity in their cabling.

The state of the sector

Thessin begins by giving her assessment of the market. She tells us, “I do worry a bit that we're sleepwalking into a crisis. We're already seeing some super low vacancy rates in key data center markets, and we have power infrastructure bottlenecks. We've got a shortage of skilled workers, and on top of that, supply chain issues continue to slowly emerge. I don't think we're acting fast enough.

“We need massive investments in infrastructure, especially renewables. We also need to train a new generation of data center professionals. Most importantly, policymakers need to understand that data infrastructure is just as important as roads and bridges. The bottom line is that awareness is growing, but I'm not sure it's growing fast enough, given the scale of what we’re up against.”

While this could be perceived as a bleak assessment, it’s important to remember that inroads are being made to resolve all these issues, and AFL is playing a significant part through education programs, webinars, work placements, and collaborations with groups such as the not-for-profit 7x24 Exchange, an education forum for the industry, of which Thessin herself is a member.

The challenge of staying connected

Turning to the subject of interconnects, Thessin points out that cabling is what allows tens of thousands of processors to handle trillions of parameters. She tells us, “I think that data center operators cannot afford to overlook their interconnects. Cooling and power are important, but the network is the backbone that enables AI at scale. It's a massive challenge.”

With each high-end GPU requiring a 400-gigabit Ethernet interface and 800-gigabit interfaces beginning to emerge in hyperscale environments, that’s a lot of cable – but it’s a vital component for delivering modern workloads at scale. But, even with the right capacity of fiber connection, the way the data center is connected plays a vital role in maximizing throughput. As Thessin explains, the key is less spaghetti, and more structure: “Organizing your cables into standardized pathways, helps you to simplify management and allows you to pack more connections into the same space.”

Then, there’s choosing the right connectors: “Pre-terminated cables can help minimize connector loss, while also saving you a bunch of installation time, and then for cross-connects, having dedicated patching areas makes your network a lot more flexible. What's most important is having a cabling strategy that helps future-proof your data center. Installing a backbone that can handle multiple generations of equipment and iterations allows you to adapt to those increasing speeds without having to rip and replace every time that you have a technology upgrade.”

Hack your cables

A surprisingly common issue that many data centers overlook is that of cable management. Poor cabling practices can lead to poor airflow, which decreases the efficiency of cooling, and in turn, uses more power, raising costs. Thessin has a six-step hack that every operator can implement today, at little or no cost, which has the potential to create massive efficiencies and cost savings:

1. Label that cable: “This can dramatically reduce your troubleshooting time and prevent errors during changes or upgrades. Using a consistent labeling scheme ensures that everybody can follow it.”
2. Tidy up: “Take the time to clean up your existing cabling, and remove anything unused or dead, because that adds to your impeding airflow. Organize your cables, put some Velcro straps around them, and bundle your cables neatly.”
3. Accessorize: “Properly utilize cable management accessories, like vertical and horizontal cable management managers. Use blanking panels to cover empty rack spaces, and seal gaps in your raised floor. It makes such a significant difference, and it's relatively inexpensive to do.”
4. Be precise: “Using precise custom-length cables or pre-terminated solutions can make a significant difference in reducing cable congestion, and is essential in maintaining cooling efficiency in high-density racks.”
5. Keep it clean: “Minimize contamination on fiber end phases. Even tiny specks of dust and debris can mess with performance. It's important to stick to a strict routine of inspecting, cleaning, and then re-inspecting your connections.”
6. Straighten up: “Avoid excessively bending the fibers. Bend-insensitive fiber types are available now, but you still need to keep in mind the recommended bend radii to ensure that your cables are performing as intended.”

Her overarching message is to think beyond simple housekeeping: “A lot of folks think that keeping things clean and neat is just about aesthetics, but it's more than that. It's a crucial strategy to maximize the efficiency, performance, and scalability of your network. Think of it as creating that ideal environment for your data center's nervous system.”

Thessin notes that the Telecommunications Industry Association (TIA) has a standard for cable labeling and management, but also concedes that due to the sheer size and variation in individual facilities, it can only ever be a starting point. Consistency is key, however, so it is important to design a scheme that will ensure a common standard across all of your facilities.

Next generation fiber

But what of next-generation fiber cabling? For a long time, a single glass fiber measured a minimum of 250 microns in diameter. Today, 200 micrometers is becoming more common and companies, including AFL, are working to create even thinner strands. These smaller fibers allow more to be bunched into a single cable, enabling greater throughput. The company currently offers Ultra High Density (UHD) cabling with 6912 fibers packed into a single cable – something that would have been unimaginable a few years ago.

The advantages are myriad, as Thessin explains: “It allows you to severely, and extensively, expand network capacity without increasing the physical space required for cabling. When you can pack 1000s of fibers into a single cable, you can support much higher bandwidth and ultimately future-proof your network against growing demand, while the high density of those fibers means that your data center can achieve the same or greater network capacity in a much smaller physical space.“

There are other bonuses too: “It reduces the cable management overhead, lowers your cooling requirements, and improves your airflow. Ultimately, it also lets you easily scale up bandwidth over time without major infrastructure changes. If you put a higher fiber count in place, and you don't light all of the fibers on day one, it gives you a pay-as-you-grow approach. You can add more fiber capacity incrementally as it's needed, which is more cost-effective.”

Hurry up and wait?

But just as consumers have waited for the next iteration of their favorite smartphone, the advent of ever-denser cabling could lead operators to risk waiting for the next upgrade. Thessin warns against this, assuring us that today’s cables are already pretty future-proofed:

“We have researchers and certainly there are companies already deploying new technologies like multi-core fiber and hollow-core fiber that can potentially provide even greater capacity and lower latency in the future. But with that said, I wouldn't recommend that operators just hold out for the next iteration. The current cables are more than capable of meeting the near-term demands, as well as providing a solid foundation for scaling up as demand increases.”

She also points out that waiting out the next big thing can put you at a competitive disadvantage because operators need to be working with 400G and 800G throughputs. After all, those that don’t are likely to lose out to those that do. The need is out there today, and the risk is low because just as is happening with the modular migration to liquid cooling, the move beyond 800G is likely to be gradual, and equally modular.

She explains, “Timelines to commercialize new emerging technologies can be unpredictable. It may be years before we see widespread adoption, and while just having cable technology is one thing, you also need that ecosystem. You need testers, splicing, and equipment connectivity to bring those new technologies to market. My advice is to invest in the current UHD cable solution, but do so with a flexible strategy in mind that allows you to perform future upgrades as needed.“

Deus ex machina

While the focus for operators must be to service the huge amounts of traffic generated by AI, it is often forgotten that machine learning (ML) has applications within the data center environment, too. Thessin is enthused at this prospect as she outlines some of the possibilities:

“This is where it gets exciting, right? We’re already seeing many benefits to AI, such as predictive maintenance where AI algorithms can analyze sensor data to predict equipment failures before they happen. That's significant. It helps you reduce downtime and extends your hardware lifespan.”

There are also potential benefits for energy efficiency: “AI can help optimize your cooling systems, adjusting real-time-based workloads and environmental conditions. Google recently announced that they were able to reduce their cooling costs by up to 40 percent by leveraging AI.”

Thessin also points to capacity planning, whereby historical data and trends can be crunched by AI to predict future needs, taking into account the accelerated rates of demand being seen across the industry.

However, she believes that some of the most exciting use cases come from our topic today – network optimization: “You can dynamically let AI route traffic balance your workloads, and reconfigure network topologies. It maximizes performance and helps you minimize latency, which is critical. I find it very intriguing to leverage AI to optimize fiber optics,” adding that in terms of infrastructure, there is significant research being conducted into AI for fiber inspection, fault detecting, and calculating future fiber requirements:

“There's a lot of potential in leveraging AI because, at the end of the day, we have many data points that are based on our existing infrastructure that we can leverage to make sure that we're making the right decisions and investments.“

AI is just one area where AFL has forged partnerships to help its customers optimize the data center. Another example of where choosing a great partner is essential is in finding the right interconnects for your fiber cabling. The rise of very small form factor (VSFF) connectors has allowed increased density to patch panels and server racks, but the term refers to a class of connectors, rather than a single universal standard. Choosing the right one for you can be a minefield.

How to be very small

Thessin advises on best practices for equipping your facilities: “It can be tricky. There are many VSFF connectors on the market. They all have their benefits. The key here is to be strategic in your approach. Take time to carefully evaluate each connector type, based on your specific needs, like your density requirements, compatibility with your current setup, and plans for your future scalability. Don't just jump on the bandwagon of the latest trend without thinking it through, there's a risk of getting locked into a vendor, especially if you go all in on a single, proprietary solution.”

Although organizations such as TIA are working to create a common standard for VSFF and reduce the levels of fragmentation currently emerging, this is some way off, so Thessin has some more top tips:

1. Adopt a modular approach: “You can mix and match different connector types within the same system, and adapt as standards change.”
2. Work with multiple vendors: “Avoid becoming time-tied to a single vendor – keep in mind we’re in a transition period.”
3. Be flexible: “The Lucent connector (LC) is still widely used. Data centers should take a hybrid approach, with VSFF for new high-density areas and keeping LC in other places.”

Edge of heaven

Up until now, we’ve spoken in terms of main data center facilities, but the conversation turns to serving Edge facilities, which are increasingly being used to offer low latency and higher coverage in even the most remote areas. Thessin reminds us that the quality of Edge connections needs to be just as good, if not better, than those in the traditional data center white space:

“It's very important to recognize that some Edge sites often face harsh environmental conditions. Think about using ruggedized fiber optic cables and connectors that are specifically designed for outdoor use. Those components are built to withstand temperature fluctuations, for example, moisture and physical stress.”

Another important consideration for Edge sites is redundancy because connections can be more vulnerable to failure, and more difficult to access for maintenance.

She adds, “They also need easy upgrades down the road, we're seeing a lot of operators install higher fiber count cables to ensure that they can do future expansions without major re-cabling efforts, because, having to work with right of way or limited duct space becomes costly and disruptive when you have to free cable down the road.”

Green glass

Finally, we turn to the subject of sustainability. After all, if you’re stripping out miles of redundant fiber to fit out facilities for the future, there can be a lot to consider to minimize waste and environmental impact. Thessin speaks to this major aspect of the process: “We have solutions to help increase the lifespan of the fiber optic solutions, and there's also innovation in the packaging, and then AFL’s manufacturing process to ensure that we're helping our customers meet their sustainability and carbon neutral goals.“

But what about those old cables? Can they be recycled?

“Recycling is always an option, but one that is not always widely available or feasible due to the complexities of the process. What we focus on is extending the life of the cables, so you do not have to recycle them frequently, but fiber optic cables are constructed of recyclable materials – whether it's the sheathing and then ultimately the fiber itself.“

Partnerships matter

Working with AFL doesn’t stop with a sale. If any of Thessin’s advice sounds mind-boggling, they’ll be there to help.

She tells us, “AFL is very much focused on the customer first. It starts from the initial engagement, where we're helping assess their needs. At the end of the day, we want to help solve their challenges. Post installation, we work with the installer partners to ensure they're trained, so we have field engineers who work closely with the installation teams and customers..

“We are always dispatching folks on-site to assess any post-installation needs and provide guidance to ensure that the investment that our clients have made in AFL infrastructure is ultimately an investment that will produce the right results.”

For more information click here