Meta's data centers underpin a vast array of Internet services and have faced unprecedented demand due to the rapid expansion of AI workloads. The traditional approach of building standardized data centers is increasingly challenged by the exponential growth in required capacity that is now sourced in a variety of non-standard physical environments and data center designs. This shift introduces a complex challenge: how to rapidly and repeatably design custom data center networks that balance multiple, often conflicting, objectives across diverse engineering disciplines.

Meta's rapid expansion in users, business operations, and AI workloads is straining our backbone network, while physical constraints—such as fiber, space, and power—limit the speed of capacity growth. To address these challenges, we present a service-aware network capacity planning suite that systematically improves network efficiency with a service collaboration approach. We propose the "safe capacity" abstraction which enables services to incorporate current and projected network conditions into their compute and storage allocation decisions. We introduce a hose-carving method that efficiently translates service-level traffic demands into detailed traffic matrices, allowing for more precise bandwidth allocation. To promote responsible network usage, we design a network rate card which attributes network consumption to individual services, incentivizing optimization and resource trade-offs. Additionally, new enforcement features at the end-host layer dynamically adjust resource allocations and traffic flows at runtime to maximize utilization. This paper is the first to detail a collaborative, service-aware approach to backbone network efficiency at Meta scale. Based on years of operational experience, we share practical insights and highlight new directions for research in network efficiency.

Backbone networks operated by cloud providers serving AI training clusters face a growing disconnect between IP-layer capacity planning and optical-layer physical constraints: IP planners decide how much capacity to turn up, but without visibility into spectrum fragmentation, channel placement, ROADM connectivity, and per-band heterogeneity introduced by multiband (C+L) transmission, they produce plans that require additional iterations to converge at the optical layer. Solving the joint IP-optical optimization is computationally prohibitive at production scale. We present FABRIC (Fragmentation-Aware Backbone Resource and IP Capacity planner), a cross-layer spectrum planning system that models spectrum at the lowest frequency-slot granularity, groups IP links by shared-risk link group, computes fragmentation losses from channel data, and feeds these constraints into a weekly capacity optimization cycle. We evaluate FABRIC and show that spectrum-aware planning accurately produces capacity plans grounded in optical-layer reality.

With the evolution of modern telecommunications, the fronthaul network has become an indispensable component of the infrastructure, particularly in the context of Cloud Ra-dio Access Networks. Nevertheless, fronthaul networks still face significant challenges such as power outages, particularly when a disaster, like an earthquake or severe weather event, occurs. Damage to power supply facilities may cause operational failures while communication is one of the most crucial needs in the disaster area to make rescue operations more effective. Specialized fibers that can deliver electrical power can help mitigate this problem. However, power losses may be extremely large over distances and there is no flexibility after the installation of the fibers. The network topology design that reduces capital and operational costs while satisfying power constraints is an important problem. In this paper, we focus on network topology design in an urban area by taking into account both fiber and power costs. We then propose integer linear programming and fast algorithms based on methods for single-facility location problems. The results demonstrate that multiple approaches help to achieve the optimal design, with our proposed method standing out due to its efficiency in finding feasible solutions, scalability, and ≈656x reduction in execution time.

C-band enabled Elastic optical networks (EONs) have been one of the most deployed optical network solutions in the world. However, as traffic demands continue to increase, capacity exhaustion is inevitable. There are two major technologies, namely, multiband elastic optical networks (MB-EONs) and space division multiplexed elastic optical networks (SDM-EONs) that can enhance capacity. Each technology offers a tradeoff between better capacity and deployment overhead which directly affects the network performance. Considering the different characteristics of these two technologies, we present our proposed strategy, Progressive Optics Deployment and Integration for Growing Yields (PRODIGY), to gradually migrate the current C-band EONs. PRODIGY uses various proactive measures, inspired by Swiss Cheese Model, to make the network robust for handling network traffic peaks and ensure that the service level agreement is met. We present a detailed comparison of our proposed strategy with customized baseline strategies, and demonstrate the superiority of our proposed approach.

Routing and Spectrum Assignment (RSA) is a well-known problem in Elastic Optical Networks (EONs), and the literature abounds with numerous heuristic algorithms that attempt to efficiently utilize spectrum and mitigate spectrum fragmentation. However, blocking performance modeling of EONs is a relatively under-explored topic, mainly because of the complexity of handling variable request sizes. In this paper, we analyze three spectrum assignment policies and compare their blocking performance. In the full slicing policy, every request is sliced into multiple one-slot calls. In the partitioning policy, each request size is allocated a dedicated spectrum partition, and each request can be only assigned spectrum from its corresponding partition. The partition with slicing policy is a hybrid of the two – there are dedicated partitions for small request sizes, but large requests are sliced into multiple sub-requests of the smaller sizes and assigned spectrum from those partitions. We obtain expressions for the bandwidth blocking probability for these three policies, and validate the analysis with extensive simulation results. Our results show that partitioning with slicing has the best performance.

Recent advances in Spectrally-Spatially Flexible Optical Networks (SS-FONs) have enhanced service provisioning by leveraging elastic optical transmission with a fine-grained and flexible frequency grid, multiple spatial modes (fiber cores), and a variety of advanced modulation formats (MFs) to increase the capacity of optical fiber. An important lightpath resource assignment problem in SS-FONs is the routing, modulation, core, and spectrum assignment (RMCSA) problem. Crosstalk (XT) between connections on different cores degrades the quality of transmission, and the RMCSA algorithm must ensure that XT constraints are met while maximizing performance. In this paper, we propose an RMCSA algorithm called Tridental Resource Assignment algorithm (TRA), as it balances network capacity, spectrum utilization and spectrum fragmentation that affect network performance. Our resource assignment approach includes both an offline/static network planning component and an online/dynamic provisioning component. In the former, MFs and spectrum utilization are used to compute path priorities for a lightpath. The dynamic provisioning component then allocates the resources on a selected path using TRA. Extensive simulation experiments performed in realistic network scenarios indicate that TRA significantly reduces the bandwidth blocking probability (BBP) in a variety of scenarios by maintaining a good balance between spectrum utilization and XT.

In traditional networks, the pre-routed packets are dropped during the link failure leading to a huge data loss. Survivability techniques such as protection and restoration are available to provide the solution before and after the link failure. But, the new flow entries that are to be added in the flow table increase the initial network demand leading to an increase in memory demand per switch. The saved data not only reduces the network speed but also demands of repeated processing of entries. In this paper, we propose a new scheme of Group Table based Rerouting (GTR) technique to find the response against single link failure through Fast Fail-over (FF) group table feature provided by OpenFlow. This scheme provides equal roles for both controllers and forwarding OpenFlow enabled switches. Here, the controller maintains a look-up table which is updated periodically according to the change in network structure. Also, it has to update the FF group table simultaneously corresponding to every active port of the switches. The controller relabels the packets and updates the flow entries on respective switches.

Software Defined Networking (SDN) takes out the control plane of the network elements and distributes it to separate entities named as controllers. The main aim of the Controller Placement Problem (CPP) is to compute the optimum locations for controller deployment and assignment of switches to respective controllers. Every controller failure results in disconnection of switches from the controller leading to reassignment of switches to one or more of the active controllers. In this paper, we propose a controller placement strategy for multiple link failures, considering the network properties, to minimize the worst case latencies. We propose the solution using decision making under uncertainty (and non-competitive situations). Our objective is to find the optimum positions in the network for controller placement with reduced Maximum Worst Case Latency (MWCL) and maximum utilization of the controller. In addition, we have proposed the constraint-based solution for reassignment of switches to active controllers in the network. We have evaluated the proposed methodologies on various network topologies from Internet Topology Zoo. It is found that the decision making under uncertainty provides a better solution on CPP as compared to the existing mathematical methods for any SDN network topology.