Congestion Control (CC) plays a fundamental role in optimizing traffic in Datacenter Networks (DCNs). Currently, DCNs implement two main CC protocols: DCTCP and DCQCN. Both protocols are based on Explicit Congestion Notification (ECN), where switches mark packets when they detect congestion. Nowadays, network experts carefully set ECN parameters to optimize the average network performance. However, today’s DCNs experience rapid and abrupt changes that severely affect the network state (e.g., dynamic workloads, incasts), which leads to under-utilization and sub-optimal performance. In this paper we present GraphCC , a framework for in-network CC optimization. GraphCC relies on Multi-agent Reinforcement Learning (MARL) and Graph Neural Networks (GNN), and is compatible with widely deployed ECN-based CC protocols. The proposed solution deploys distributed agents on switches that communicate with their neighbors to cooperate and optimize the global ECN configuration. In our evaluation, we test GraphCC with three real-world traffic workloads, focusing on its capability to accommodate scenarios unseen during training (e.g., traffic changes, failures). We compare GraphCC with a state-of-the-art MARL solution for ECN tuning, and observe that our method outperforms the state-of-the-art baseline in all evaluation scenarios, with improvements up to 20% in average Flow Completion Time, similar mean throughput (within 1%), and significant reductions in buffer occupancy (38.0–85.7%).

Network models are an essential block of modern networks. For example, they are widely used in network planning and optimization. However, as networks increase in scale and complexity, some models present limitations, such as the assumption of Markovian traffic in queuing theory models, or the high computational cost of network simulators. Recent advances in machine learning, such as Graph Neural Networks (GNN), are enabling a new generation of network models that are data-driven and can learn complex non-linear behaviors. In this paper, we present RouteNet-Fermi, a custom GNN model that shares the same goals as Queuing Theory, while being considerably more accurate in the presence of realistic traffic models. The proposed model predicts accurately the delay, jitter, and packet loss of a network. We have tested RouteNet-Fermi in networks of increasing size (up to 300 nodes), including samples with mixed traffic profiles — e.g., with complex non-Markovian models — and arbitrary routing and queue scheduling configurations. Our experimental results show that RouteNet-Fermi achieves similar accuracy as computationally-expensive packet-level simulators and scales accurately to larger networks. Our model produces delay estimates with a mean relative error of 6.24% when applied to a test dataset of 1,000 samples, including network topologies one order of magnitude larger than those seen during training. Finally, we have also evaluated RouteNet-Fermi with measurements from a physical testbed and packet traces from a real-life network.

Network modeling is an essential tool for network planning and management. It allows network administrators to explore the performance of new protocols, mechanisms, or optimal configurations without the need for testing them in real production networks. Recently, Graph Neural Networks (GNNs) have emerged as a practical solution to produce network models that can learn and extract complex patterns from real data without making any assumptions. However, state-of-the-art GNN-based network models only work with traffic matrices, this is a very coarse and simplified representation of network traffic. Although this assumption has shown to work well in certain use-cases, it is a limiting factor because, in practice, networks operate with flows. In this paper, we present FlowDT a new DL-based solution designed to model computer networks at the fine-grained flow level. In our evaluation, we show how FlowDT can accurately predict relevant per-flow performance metrics with an error of 3.5%, FlowDT’s performance is also benchmarked against vanilla DL models as well as with Queuing Theory.

Network modeling is a fundamental tool in network research, design, and operation. Arguably the most popular method for modeling is Queuing Theory (QT). Its main limitation is that it imposes strong assumptions on the packet arrival process, which typically do not hold in real networks. In the field of Deep Learning, Graph Neural Networks (GNN) have emerged as a new technique to build data-driven models that can learn complex and non-linear behavior. In this paper, we present RouteNet-Erlang, a pioneering GNN architecture designed to model computer networks. RouteNet-Erlang supports complex traffic models, multi-queue scheduling policies, routing policies and can provide accurate estimates in networks not seen in the training phase. We benchmark RouteNet-Erlang against a state-of-the-art QT model, and our results show that it outperforms QT in all the network scenarios.

This paper introduces Traffic Encoding Matrices (TEMs) as an alternative to traditional Traffic Matrices (TMs) for representing network traffic demands. While TMs only capture average demand information, TEMs are designed to capture distributional demand information by learning representations whose variations capture most of the structure of the distribution of demands. We present a practical approach based on off-the-shelf neural autoencoders to efficiently construct TEMs at the edge of the network. We then present the design and evaluation of NeuroTE, a DRL-based framework for delay-aware traffic engineering in SD-WANs using TEMs. Using real traffic traces, we present experimental results to demonstrate the advantages of using TEMs instead of TMs in traffic engineering. Our results show that, when traffic demands are dynamic, TEM-based control leads to: 1) improved network performance, and 2) faster convergence to the optimal solution compared to TM-based control using exactly the same DRL control algorithm.

The recent growth of emergent network applications (e.g., satellite networks, vehicular networks) is increasing the complexity of managing modern communication networks. As a result, the community proposed the Digital Twin Networks (DTN) as a key enabler of efficient network management. Network operators can leverage the DTN to perform different optimization tasks (e.g., Traffic Engineering, Network Planning).Deep Reinforcement Learning (DRL) showed a high performance when applied to solve network optimization problems. In the context of DTN, DRL can be leveraged to solve optimization problems without directly impacting the real-world network behavior. However, DRL scales poorly with the problem size and complexity. In this paper, we explore the use of Evolutionary Strategies (ES) to train DRL agents for solving a routing optimization problem. The experimental results show that ES achieved a training time speed-up of 128 and 6 for the NSFNET and GEANT2 topologies respectively.

Wide Area Networks (WAN) are a key infrastructure in today’s society. During the last years, WANs have seen a considerable increase in network’s traffic and network applications, imposing new requirements on existing network technologies (e.g., low latency and high throughput). Consequently, Internet Service Providers (ISP) are under pressure to ensure the customer’s Quality of Service and fulfill Service Level Agreements. Network operators leverage Traffic Engineering (TE) techniques to efficiently manage the network’s resources. However, WAN’s traffic can drastically change during time and the connectivity can be affected due to external factors (e.g., link failures). Therefore, TE solutions must be able to adapt to dynamic scenarios in real-time. In this paper we propose Enero, an efficient real-time TE solution based on a two-stage optimization process. In the first one, Enero leverages Deep Reinforcement Learning (DRL) to optimize the routing configuration by generating a long-term TE strategy. To enable efficient operation over dynamic network scenarios (e.g., when link failures occur), we integrated a Graph Neural Network into the DRL agent. In the second stage, Enero uses a Local Search algorithm to improve DRL’s solution without adding computational overhead to the optimization process. The experimental results indicate that Enero is able to operate in real-world dynamic network topologies in 4.5 s on average for topologies up to 100 links.

Network modeling is a critical component of Quality of Service (QoS) optimization. Current networks implement Service Level Agreements (SLA) by careful configuration of both routing and queue scheduling policies. However, existing modeling techniques are not able to produce accurate estimates of relevant SLA metrics, such as delay or jitter, in networks with complex QoS-aware queueing policies (e.g., strict priority, Weighted Fair Queueing, Deficit Round Robin). Recently, Graph Neural Networks (GNNs) have become a powerful tool to model networks since they are specifically designed to work with graph-structured data. In this paper, we propose a GNN-based network model able to understand the complex relationship between (i) the queueing policy (scheduling algorithm and queue sizes), (ii) the network topology, (iii) the routing configuration, and (iv) the input traffic matrix. We call our model TwinNet, a Digital Twin that can accurately estimate relevant SLA metrics for network optimization. TwinNet can generalize to its input parameters, operating successfully in topologies, routing, and queueing configurations never seen during training. We evaluate TwinNet over a wide variety of scenarios with synthetic traffic and validate it with real traffic traces. Our results show that TwinNet can provide accurate estimates of end-to-end path delays in 106 unseen real-world topologies, under different queuing configurations with a Mean Absolute Percentage Error (MAPE) of 3.8%, as well as a MAPE of 6.3% error when evaluated with a real testbed. We also showcase the potential of the proposed model for SLA-driven network optimization and what-if analysis.

The digital transformation is pushing the existing network technologies towards new horizons, enabling new applications (e.g., vehicular networks). As a result, the networking community has seen a noticeable increase in the requirements of emerging network applications. One main open challenge is the need to accommodate control systems to highly dynamic network scenarios. Nowadays, existing network optimization technologies do not meet the needed requirements to effectively operate in real time. Some of them are based on hand-crafted heuristics with limited performance and adaptability, while some technologies use optimizers which are often too time-consuming. Recent advances in Deep Reinforcement Learning (DRL) have shown a dramatic improvement in decision-making and automated control problems. Consequently, DRL represents a promising technique to efficiently solve a variety of relevant network optimization problems, such as online routing. In this paper, we explore the use of state-of-the-art DRL technologies for real-time routing optimization and outline some relevant open challenges to achieve production-ready DRL-based solutions.

Traffic Engineering (TE) is a basic building block of the Internet. In this paper, we analyze whether modern Machine Learning (ML) methods are ready to be used for TE optimization. We address this open question through a comparative analysis between the state of the art in ML and the state of the art in TE. To this end, we first present a novel distributed system for TE that leverages the latest advancements in ML. Our system implements a novel architecture that combines Multi-Agent Reinforcement Learning (MARL) and Graph Neural Networks (GNN) to minimize network congestion. In our evaluation, we compare our MARL+GNN system with DEFO, a network optimizer based on Constraint Programming that represents the state of the art in TE. Our experimental results show that the proposed MARL+GNN solution achieves equivalent performance to DEFO in a wide variety of network scenarios including three real-world network topologies. At the same time, we show that MARL+GNN can achieve significant reductions in execution time (from the scale of minutes with DEFO to a few seconds with our solution).

With the quick growth of traffic between data centers, inefficient transfer scheduling in inter-datacenter networks can lead to a huge waste of bandwidth thus significant bandwidth cost. Previous work have explored different ways, such as software-defined WANs and dynamic pricing mechanisms, to overcome the inefficiency of inter-datacenter networks. However, there is a big challenge in addressing the fundamental conflicts between the deadline-aware transfer scheduling and minimizing the bandwidth cost. Unlike existing efforts that schedule inter-datacenter transfers under fixed link capacities, wherein some deadlines are violated and the service quality is degraded, we aim to finish all the transfers on time with as little bandwidth as possible to minimize the bandwidth cost. We take into account the variation of bandwidth price and the deadline requirements of services, and formulate the problem of cost-efficient scheduling of bulk transfers with deadline guarantee, which is shown to be NP-hard. Benefitting from the relax-and-round method, we propose a progressively-descending algorithm (PDA) to schedule bulk transfers and meet the above goals with a guaranteed approximation ratio. We apply our algorithm in a bulk transfer scheduler, Butler, and build a small-scale testbed to evaluate its efficiency. Both large-scale simulation and testbed experiment results validate the ability of our scheme on cutting down the bandwidth cost. Compared with existing approaches, it reduces up to 60% bandwidth cost and increases the network utilization by up to 140%.

The introduction of wireless transmissions into the data center has shown to be promising in improving cost effectiveness of data center networks (DCNs). For high transmission flexibility and performance, a fundamental challenge is to increase the wireless availability and enable fully hybrid and seamless transmissions over both wired and wireless DCN components. Rather than limiting the number of wireless radios by the size of top-of-rack switches, we propose a novel DCN architecture, Diamond, which nests the wired DCN with radios equipped on all servers. To harvest the gain allowed by the rich reconfigurable wireless resources, we propose the low-cost deployment of scalable 3-D ring reflection spaces (RRSs) which are interconnected with streamlined wired herringbone to enable large number of concurrent wireless transmissions through high-performance multi-reflection of radio signals over metal. To increase the number of concurrent wireless transmissions within each RRS, we propose a precise reflection method to reduce the wireless interference. We build a 60-GHz-based testbed to demonstrate the function and transmission ability of our proposed architecture. We further perform extensive simulations to show the significant performance gain of diamond, in supporting up to five times higher server-to-server capacity, enabling network-wide load balancing, and ensuring high fault tolerance.

Energy consumption has become a key issue for running large-scale data center networks (DCN) nowadays. Previous studies mainly focus on energy saving through reducing the number of active servers or network switches with traffic consolidation. However, since this mechanism benefits from the routing flexibility, both the gained energy savings and flow performance are limited by the conventional static network topology. Recent advances in DCN architecture propose to implement an adaptive network topology with reconfigurable optical/wireless links (i.e., topology-adaptive DCNs), which has shown a great potential to improve the transmission performance of existing static wired network. In this paper, we propose GreenWay, an energy-efficient solution to jointly optimize the virtual machine (VM) placement and flow transmissions under the new paradigm of topology adaption. Based on the VM traffic demands, it can construct a proper run-time network topology to lower both the energy cost and communication cost among VMs. We first formulate the energy optimization problem in topology-adaptive DCNs. After showing the its NP-hardness, we then develop heuristic algorithms to address the VM placement and topology adaption effectively. Extensive trace-based simulations show that GreenWay consumes much less energy than other state-of-the- art solutions while ensuring better flow performance. We finally implement an OpenvSwitch- based testbed and demonstrate the efficiency of our solution.

Multi-path TCP has recently shown great potential to take advantage of the rich path diversity in data center networks (DCN) to increase transmission throughput. However, the small flows, which take a large fraction of data center traffic, will easily get a timeout when split onto multiple paths. Moreover, the dynamic congestions and node failures in DCN will exacerbate the reorder problem of parallel multi-path transmissions for large flows. In this paper, we propose DC2-MTCP (Data Center Coded Multi-path TCP), which employs a fast and light-weight coding method to address the above challenges while maintaining the benefit of parallel multi-path transmissions. To meet the high flow performance in DCN, we insert a very low ratio of coded packets with a careful selection of the packets to be coded. We further present a progressive decoding algorithm to decode the packets online with a low time complexity. Extensive ns2-based simulations show that with two orders of magnitude lower coding delay, DC2-MTCP can reduce on average 40% flow completion time for small flows and increase 30% flow throughput for large flows compared to the peer schemes in varying network conditions.

In cellular networks, it is important to conserve energy while at the same time satisfying different user performance requirements. In this paper, we first propose a comprehensive metric to capture the user performance cost due to task delay, deadline violation, different application profiles, and user preferences. We prove that finding the energy-optimal scheduling solution while meeting the requirements on the performance cost is NP-hard. Then, we design an adaptive online scheduling algorithm PerES to minimize the total energy cost on data transmissions subject to user performance constraints. We prove that PerES can make the energy consumption arbitrarily close to that of the optimal scheduling solution. Further, we develop offline algorithms to serve as the evaluation benchmark for PerES. The evaluation results demonstrate that PerES achieves average 2.5 times faster convergence speed compared to state-of-art static methods, and also higher performance than peers under various test conditions. Using 821 million traffic flows collected from a commercial cellular carrier, we verify our scheme could achieve on average 32-56 percent energy savings over the total transmission energy with different levels of user experience.

Virtual machine (VM) migration is a key technique for network resource optimization in modern data center networks. Previous work generally focuses on how to place the VMs efficiently in a static network topology by migrating the VMs with large traffic demands to close servers. As the flow demands between VMs change, however, a great cost will be paid for the VM migration. In this paper, we propose a new paradigm for VM migration by dynamically constructing adaptive topologies based on the VM demands to lower the cost of both VM migration and communication. We formulate the traffic-aware VM migration problem in an adaptive topology and show its NP-hardness. For periodic traffic, we develop a novel progressive-decompose-rounding algorithm to schedule VM migration in polynomial time with a proved approximation ratio. For highly dynamic flows, we design an online decision-maker (ODM) algorithm with proved performance bound. Extensive trace-based simulations show that PDR and ODM can achieve about four times flow throughput among VMs with less than a quarter of the migration cost compared to other state-of-art VM migration solutions. We finally implement an OpenvSwitch-based testbed and demonstrate the efficiency of our solutions.

Virtual machine (VM) migration is a key technique for network resource optimization in modern data center networks (DCNs). Previous work generally focuses on how to place the VMs efficiently in a static network topology by migrating the VMs with large traffic demands to close servers. When the VM demands change, however, a great cost will be paid on the VM migration. With the advance of software-defined network (SDN), recent studies have shown great potential to implement an adaptive network topology at a low cost. Taking advantage of the topology adaptability, in this paper, we propose a new paradigm for VM migration by dynamically constructing a topology based on the VM demands to lower the cost of both VM migration and communication. We formulate the traffic-aware VM migration problem in an adaptive topology and show its NP-hardness. Then we develop a novel progressive-decompose-rounding (PDR) algorithm to solve this problem in polynomial time with a proved approximation ratio. Extensive trace-based simulations show that PDR can achieve higher flow throughput among VMs with only a quarter of the migration cost compared to other state-of-art VM migration solutions. We finally implement an OpenvSwitch-based testbed and demonstrate the efficiency of our solution.

The densely deployed Access Points (APs) have overlapping coverage areas. In the hotspot area, a user can usually receive signals of more than ten APs. The signal-based association in IEEE 802.11 may result in significant unbalanced loads among APs. Moreover, diverse user demands on bandwidth further exacerbate the load unbalance. Some APs are too overloaded when multiple high-demand users gather on them due to the strongest signal, while the others are light-loaded with a few low-demand users associated to them. The severe load unbalance degrades the user performance. In this paper, by adding different user bandwidth demands as new constraints, we formulate the joint AP association and bandwidth allocation problem. We comprehensively analyze the solution space of optimal bandwidth allocation while satisfying the time-based fairness among users. We develop a 1/2-approximation algorithm to solve the problem. Our extensive trace- driven evaluations show that our algorithm achieves better load balance. As a result, it greatly improves the aggregated throughput and provides better user fairness than conventional association schemes.

Energy consumption of data center has become an important challenge due to high electric cost and carbon dioxide emissions. Previous work has mainly focused on saving energy cost of servers, though the energy consumption of data center networks (DCNs), consisting of networking equipments like switches, also takes a significant part of the overall energy consumption. In this paper, we propose ProCons, an energy saving mechanism that dynamically consolidates traffic flows onto a small set of networking equipments in order to shut down idle ones for energy saving. Different from previous works that assume the traffic demands to be stable, ProCons takes into account the variance of traffic demand over time, and predicts future demand based on historical statistics. The traffic flows are then scheduled based on the predicted future demands and the capacity of each link. We evaluate ProCons with real life traces collected from data centers using a flow-level simulator. Our experimental results show that using ProCons, 40% of energy savings for DCNs can be gained while maintaining the good performance of flow transmission.

In cellular networks, it is important to conserve energy while at the same time ensuring users to have good transmission experiences. The energy cost can result from tail energy due to the radio resource control strategies designed in cellular networks and data transmission. Existing efforts generally consider one of the energy issues, and also ignore the adverse impact on user transmission performance due to energy conservation. In addition, many existing algorithms are based on prediction and knowledge on future traffic, which are hard to apply in a practical wireless system with dynamic user traffic and channel condition. The goal of this work is to design an efficient online scheduling algorithm to minimize energy consumption both due to tail energy and transmissions while meeting user performance expectation. We prove the problem to be NP-hard, and design a practical online scheduling algorithm PerES to minimize the total energy cost of multiple mobile applications subject to user performance constraints. We propose a comprehensive performance cost metric to capture the impacts due to task delay, deadline violation, different application profiles and user preferences. We prove that our proposed scheduling algorithm can make the energy consumption arbitrarily close to that of the optimal scheduling solution. The evaluation results demonstrate the effectiveness of our scheme and its higher performance than peers. Moreover, by supporting dynamic performance requirement by mobile users, PerES can achieve 2 times faster convergence to both the performance degradation bound and optimal energy conversation bound than those of traditional static methods. Using 821 million traffic flows collected from a commercial cellular carrier, we verify our scheme could achieve on average 32%-56% energy savings with different levels of user experience.

We propose a novel architecture of data center networks (DCN), which adds wireless network card to both servers and routers. Existing traffic redundancy elimination (TRE) mechanisms reduce link loads and increase network capacity in several environments by removing strings that have appeared in earlier packets through encoding and decoding them several hops downstream. This article is the first to explore TRE mechanisms in large-scale DCNs and the first to exploit cooperative TRE among servers. Moreover, it also achieves the `logically centralized' control over the physically distributed states in emerging software defined networks (SDN) paradigm, by sharing information among servers and routers in data centers with wireless cards. We first formulate the TREDaCeN (TRE in Data Center Networks) problem and reduce the cycle cover problem to prove that finding an optimal caching task assignment for TREDaCeN problem is NP-hard. We further describe an offline TREDaCeN algorithm which is proved to have good approximation ratio. We then discuss efficient online zero-delay and semi-distributed implementations of TREDaCeN supported by physical proximity of servers and routers, enabling status updates in a single wireless transmission, using an efficient prioritized schedule. We also address online cache replacement and consistency of information in servers and routers with and without delay. Our framework is tested on different parameters and shows superior performance in comparison to other mechanisms (imported directly to this setting). Our results show the robustness and the trade-off between the `logically centralized' implementation and the overhead on handling inconsistency of distributed information in DCN.