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.

Network design is the process of dimensioning IP capacity over an optical network infrastructure to satisfy a given set of demands and reliability constraints. Specifically, we consider the problem of hose-based cross-layer network design, which seeks to find a minimum cost design that is able to route demand for all hose traffic matrices under all specified failure states. While most network design problems are solved as Mixed Integer Programs, a commercial solver can become intractable due to the scale of today's networks. We demonstrate how the classic Benders decomposition algorithm can be applied and improved for this problem and discuss practical implementation aspects. We showcase a horizontally scalable distributed framework to leverage the decomposable problem structure and solve millions of linear programs in a distributed manner, thereby making the network design problem tractable. In contrast to the conventional approach where failure states and traffic matrices are planned sequentially, the Benders algorithm finds global optimal designs across all traffic matrices and failure states. This leads to network designs with improved solution quality and reliability, with 20--30% less IP capacity and spectrum consumption, 50% less link augments and up to 20x faster runtime that enables design for hyper scale networks in a matter of hours.

In this article, we propose a new protection scheme for backbone networks to guarantee high service availability. The presented scheme does not require any reconfiguration immediately after the failure (i.e., it is proactive). At the same time, it does not require any reserved backup network resources either. To achieve these seemingly contradictory goals, we utilize the recent advancements in Machine Learning (ML) to implement a network intelligence that periodically re-allocates the unused capacity as protection bandwidth to meet the service availability requirements of each connection. Our goal is achieved by two components (1) predicting the traffic for the next period on each link, and (2) intelligently selecting the best fit dedicated protection scheme for the next period depending on the estimated unused (spare) bandwidth and the previous service availability violations. Note that re-allocating protection bandwidth affects neither the operational connections nor the current best practice of operators to over-provision network bandwidth to support elephant flows. Finally, we provide a case study on the real traffic from Energy Sciences Network (ESnet), a high-speed, international scientific backbone network. The key benefit of our framework is that adaptively utilizing the over-provisioned bandwidth for spare capacity is sufficient to improve the availability from three-nines to five-nines (in ESnet for the 30 examined connections). The drawback is negligible bandwidth limitations; the user perceives a minor and very temporal bandwidth limitation in less than 0.1% of the time.

Real-time, high-speed flow classification is fundamental for network operation tasks, including reactive and proactive traffic engineering, anomaly detection and security enhancement. Existing flow classification solutions, however, do not allow operators to classify traffic based on fine-grained, temporal dynamics due to imprecise timing, often rely on sampled data, or only work with low traffic volumes and rates. In this paper, we present Seior, a classification solution that: (i) uses precision timing, (ii) has the ability to examine every packet on the network, (iii) classifies very high traffic volumes with high precision. To achieve this, Seior exploits the data aggregation and timestamping functionality of programmable dataplanes. As a concrete example, we present how Seior can be used together with Machine Learning algorithms (such as CNN, k-NN) to provide accurate, real-time and high-speed TCP congestion control classification, separating TCP BBR from its predecessors with over 88-96% accuracy and F1-score of 0.864-0.965, while only using 15.5 MiB of memory in the dataplane.

Future networks are expected to support low-latency, context-aware and user-specific services in a highly flexible and efficient manner. One approach to support emerging use cases such as, e.g., virtual reality and in-network image processing is to introduce virtualized network functions (vNF)s at the edge of the network, placed in close proximity to the end users to reduce end-to-end latency, time-to-response, and unnecessary utilisation in the core network. While placement of vNFs has been studied before, it has so far mostly focused on reducing the utilisation of server resources (i.e., minimising the number of servers required in the network to run a specific set of vNFs), and not taking network conditions into consideration such as, e.g., end-to-end latency, the constantly changing network dynamics, or user mobility patterns. In this paper, we formulate the Edge vNF placement problem to allocate vNFs to a distributed edge infrastructure, minimising end-to-end latency from all users to their associated vNFs. We present a way to dynamically re-schedule the optimal placement of vNFs based on temporal network-wide latency fluctuations using optimal stopping theory. We then evaluate our dynamic scheduler over a simulated nation-wide backbone network using real-world ISP latency characteristics. We show that our proposed dynamic placement scheduler minimises vNF migrations compared to other schedulers (e.g., periodic and always-on scheduling of a new placement), and offers Quality of Service guarantees by not exceeding a maximum number of latency violations that can be tolerated by certain applications.

Next-generation networks are expected to support low-latency, context-aware and user-specific services in a highly flexible and efficient manner. Proposed applications include high-definition, low-latency video streaming, remote surgery, as well as applications for tactile Internet, virtual or augmented reality that demand network side data processing (such as image recognition, transformation or head/eye motion aware rendering). One approach to support these use cases is to introduce virtualized network services at the edge of the network, in close proximity of the end users to reduce end-to-end latency, time-to-response and unnecessary utilization of the core network, while providing flexibility for resource allocation. While many research projects including our previous work on Glasgow Network Functions have proposed running virtual network functions (vNF)s at the network edge, a latency-optimal placement allocation has not been presented before for the network edge and therefore the impact on user-to-vNF latency has not been investigated. In this paper, we formulate a simple vNF placement problem that minimizes end-to-end latency from users to their network functions. We have implemented the problem using Integer Linear Programming (ILP) with the Gurobi solver, and evaluated it with a real topology of a network provider. We use our solution to compare two vNF deployment scenarios over an emulation of a national backbone network: a two-tier edge deployment and a cloud-only deployment. We show that, in our example, using edge servers can deliver up to 70% improvement in user-to-vNF latency.

In modern Cloud Data Centers (DC)s, correct implementation of network policies is crucial to provide secure, efficient and high performance services for tenants. It is reported that the inefficient management of network policies accounts for 78% of DC downtime, challenged by the dynamically changing network characteristics and by the effects of dynamic Virtual Machine (VM) consolidation. While there has been significant research in policy and VM management, they have so far been treated as disjoint research problems. In this paper, we explore the simultaneous, dynamic VM and policy consolidation, and formulate the Policy-VM Consolidation (PVC) problem, which is shown to be NP-Hard. We then propose Sync, an efficient and synergistic scheme to jointly consolidate network policies and virtual machines. Extensive evaluation results and a testbed implementation of our controller show that policy and VM migration under Sync significantly reduces flow end-to-end delay by nearly 40%, and network-wide communication cost by 50% within few seconds, while adhering strictly to the requirements of network policies.

While the network edge is becoming more important for the provision of customized services in next generation mobile networks, current NFV architectures are unsuitable to meet the increasing future demand. They rely on commodity servers with resource-hungry Virtual Machines that are unable to provide the high network function density and mobility requirements necessary for upcoming wide-area and 5G networks.

Software-defined networking (SDN) is an emerging paradigm to logically centralize the network control plane and automate the configuration of individual network elements. At the same time, in cloud data centers (DCs), although network and server resources are collocated and managed by a single administrative entity, disjoint control mechanisms are used for their respective management. In this paper, we propose a unified server-network resource management for such converged information and communication technology (ICT) environments. We present a SDN-based orchestration framework for live virtual machine (VM) management that exploits temporal network information to migrate VMs and minimize the network-wide communication cost of the resulting traffic dynamics. A prototype implementation is presented, and a cloud DC testbed is used to evaluate the impact of diverse orchestration algorithms. Our live VM management has been shown to reduce the network-wide communication cost, especially for the high-cost and congestion-prone core and aggregation layers of the DC. Our results show an increase in network-wide throughput by over six times, as well as over 70% communication cost reduction by migrating less than 50% of the VMs.

OpenFlow has emerged as the de facto control protocol to implement Software-Defined Networking (SDN). In its current form, the protocol specifies a set of fields on which it matches packets to perform actions, such as forwarding, discarding or modifying specific protocol header fields at a switch. The number of match fields has increased with every version of the protocol to extend matching capabilities, however, it is still not flexible enough to match on arbitrary packet fields which limits innovation and new protocol development with OpenFlow. In this paper, we argue that a fully flexible match structure is superior to continuously extending the number of fields to match upon. We use Berkeley Packet Filters (BPF) for packet classification to provide a protocol-independent, flexible alternative to today's OpenFlow fixed match fields. We have implemented a prototype system and evaluated the performance of the proposed match scheme, with a focus on the time it takes to execute and the memory required to store different match filter specifications. Our prototype implementation demonstrates that line-rate arbitrary packet classification can be achieved with complex BPF programs.

Today's enterprise networks almost ubiquitously deploy middlebox services to improve in-network security and performance. Although virtualization of middleboxes attracts a significant attention, studies show that such implementations are still proprietary and deployed in a static manner at the boundaries of organisations, hindering open innovation. In this paper, we present an open framework to create, deploy and manage virtual network functions (NF)s in OpenFlow-enabled networks. We exploit container-based NFs to achieve low performance overhead, fast deployment and high reusability missing from today's NFV deployments. Through an SDN northbound API, NFs can be instantiated, traffic can be steered through the desired policy chain and applications can raise notifications. We demonstrate the systems operation through the development of exemplar NFs from common Operating System utility binaries, and we show that container-based NFV improves function instantiation time by up to 68% over existing hypervisor-based alternatives, and scales to one hundred co-located NFs while incurring sub-millisecond latency.