Beyond the Data Center: High-Performance Networks for AI
AI is driving a massive increase in data center workloads, with IDC predicting that data generation will grow at a compound annual growth rate (CAGR) of 40.5% through 2027. Gartner expects that the use of generative AI models will influence over 90% of organizations to pursue hybrid cloud environments through 2027. This will put a significant strain on data center interconnect (DCI) networks that transport data between distributed, hybrid, and cloud data center architectures. Traditionally, transporting data between geographically distributed data centers required leasing high-capacity circuits from service providers or investing in dedicated optical transport networks. However, DCI connections over dark fiber using coherent pluggable optics offer an innovative alternative that reduces both the cost and complexity of connecting data centers to support AI applications.
The financial burden of traditional DCI solutions
Typically, service providers use their fiber networks to offer DCI services to enterprises. These leased circuits are often 10G links or multiple 10G links that support traditional DCI applications, such as disaster recovery and enterprise services. They are also used to support 100G links for more demanding applications—such as data center mirroring and high-speed data aggregation and connections—to cloud services. The adoption of more data-intensive applications is pushing DCI capacity towards high-bandwidth 400G and 800G connections. Since the cost structure of DCI is usually based on bandwidth usage, this shift creates a new financial burden for enterprises. Costs can quickly multiply as they need more circuits to support increased capacity demands.
The advantage of leased dark fiber for DCI
Leasing dark fiber is a compelling option for enterprises to mitigate the financial impact of DCI as capacity requirements increase. Dark fiber are optical fiber cables that have been laid but are not yet “lit” with data transmissions by a service provider. By leasing these fibers, enterprises can create their own private network, gaining control over their data traffic without the recurring costs associated with bandwidth usage under traditional carrier models.
The State of Maryland is a great example of the use of dark fiber for DCI. The state’s Department of Information Technology operates a 3,300-mile network and turning to dark fiber resulted in annual savings of over $111 million, with an ROI of 871%. The benefits of leasing dark fiber and building a private network extend beyond cost savings. Dark fiber gives enterprises scalability, which is vital in the age of AI. As their data needs grow, enterprises can scale their bandwidth without waiting for their service providers to add more capacity. Dark fiber also boosts enterprise security postures by providing full control over the data traversing the network, which reduces exposure to external threats.
Integrating IP and optical networks for DCI
Traditional DCI architectures are built on a dedicated optical network layer that requires investments in optical transponders and line systems to transmit data traffic across the fiber. Enterprises can reduce the cost and complexity of DCI with Cisco Routed Optical Networking. It is a proven and standardized solution, used in hyperscale DCI environments and in service provider networks. More than 300 customers worldwide have adopted Routed Optical Networking to increase capacity while reducing energy consumption and costs.
Routed Optical Networking enables optical wavelengths to be delivered directly from the high-capacity ports in routers or switches in the data center. It simplifies the network by replacing dedicated transponders with industry-standard coherent pluggable optics, high-density routers, and simpler optical line systems. This leads to a lower-cost network design. For point-to-point DCI solutions, enterprises can deploy coherent pluggable optics with Cisco Nexus switches to collapse the switching and optical layers at Layer 2. For networks supporting more complex services, they can converge routing and optical layers onto a single IP/Multiprotocol Label Switching (IP/MPLS) network where all switching happens at Layer 3.
The benefits of Routed Optical Networking for DCI
What does all this mean to an enterprise balance sheet? Fewer components mean reduced CapEx, a smaller hardware footprint, and lower power and cooling costs. According to analysis by ACG Research, Routed Optical Networking can decrease the total cost of ownership for DCI applications by a staggering 48%, while lowering CapEx by 60% and overall environmental costs (power, cooling, and floor space) by up to 83%. But it doesn’t stop there. Routed Optical Networking also contributes to:
- Network simplification: Convergence of the IP and optical layers streamlines the planning, design, activation, management, troubleshooting, and resolution of network services.
- Automation and assurance: A single end-to-end automation platform reduces errors, improves resiliency, simplifies repairs, and accelerates the launch of new services.
- Network performance: Routed Optical Networking supports high-speed connectivity for demanding applications like AI traffic or long-distance transmission.
- Reliable transmission: Robust data transmission in the coherent pluggable optics adapts to different network and fiber infrastructure conditions for enhanced transmission reliability.
DCI for AI-ready data centers
Cisco has introduced a set of accelerated compute, network fabric, and hyperscale fabric solutions to address the data center requirements of AI—from the training of sophisticated models to the widespread use of inferencing. As the volume of data powering AI surges, the need for cost-effective and scalable solutions to interconnect data centers has never been greater. By leasing dark fiber and adopting Routed Optical Networking, enterprises can ensure that their mission-critical networks are ready to support the unprecedented demands of AI today and tomorrow.
Cisco Routed Optical Networking for DCI applications.
Share: