International audience

International audience

We focus on the deployment of large scale Edge Computing (EC) infrastructure, which requires substantial capital and operational costs. The Infrastructure Providers (InPs) are reluctant to make these investments, as the revenues from edge services, e.g., augmented reality, may be mainly captured by the respective Service Providers (SPs), and not the InPs. This is the main reason for the limited deployment of EC. We tackle this economic aspect by proposing a co-investment strategy, in which all players, i.e., one InP and multiple SPs, form a coalition to deploy an optimal amount of resources, dynamically allocate them over an investment period, and fairly share costs and revenues. The major challenge is to ensure that the coalition is stable, i.e., the co-investment is profitable for all players under uncertainty of revenues, which comes from the uncertainty of future user demand. We address this challenge with a novel stochastic coalitional game formulation, which allows us to analytically compute a lower bound on the probability that the grand coalition is stable. Numerical results show a large value for the lower bound, even with high levels of uncertainty, provided that the number of SPs is relatively small, their loads are comparable, and the investment period is long enough.

We study the placement of virtual functions aimed at minimizing the cost. We propose a novel algorithm, using ideas based on multi-armed bandits. We prove that these algorithms learn the optimal placement policy rapidly, and their regret grows at a rate at most$O(N M \sqrt{T \ln T})$while respecting the feasibility constraints with high probability, where$T$is total time slots,$M$is the number of classes of function and$N$is the number of computation nodes. We show through numerical experiments that the proposed algorithm both has good practical performance and modest computational complexity. We propose an acceleration technique that allows the algorithm to achieve good performance also in large networks where computational power is limited. Our experiments are fully reproducible, and the code is publicly available.

Edge Computing (EC) is a computational paradigm that involves deploying resources such as CPUs and GPUs near end-users, enabling low-latency applications like augmented reality and real-time gaming. However, deploying and maintaining a vast network of EC nodes is costly, which can explain its limited deployment today. A new paradigm called Vehicular Cloud Computing (VCC) has emerged and inspired interest among researchers and industry. VCC opportunistically utilizes existing and idle vehicular computational resources for external task offloading. This work is the first to systematically address the following question: Can VCC replace EC for low-latency applications? Answering this question is highly relevant for Network Operators (NOs), as VCC could eliminate costs associated with EC given that it requires no infrastructural investment. Despite its potential, no systematic study has yet explored the conditions under which VCC can effectively support low-latency applications without relying on EC. This work aims to fill that gap. Extensive simulations allow for assessing the crucial scenario factors that determine when this EC-to-VCC substitution is feasible. Considered factors are load, vehicles mobility and density, and availability. Potential for substitution is assessed based on multiple criteria, such as latency, task completion success, and cost. Vehicle mobility is simulated in SUMO, and communication in NS3 5G-LENA. The findings show that VCC can effectively replace EC for low-latency applications, except in extreme cases when the EC is still required (latency < 16 ms ).

We consider a Network Operator (NO) that owns Edge Computing (EC) resources, virtualizes them and lets third party Service Providers (SPs) run their services, using the allocated slice of resources. We focus on one specific resource, i.e., cache space, and on the problem of how to allocate it among several SPs in order to minimize the backhaul traffic. Due to confidentiality guarantees, the NO cannot observe the nature of the traffic of SPs, which is encrypted. Allocation decisions are thus challenging, since they must be taken solely based on observed monitoring information. Another challenge is that not all the traffic is cacheable. We propose a data-driven cache allocation strategy, based on Reinforcement Learning (RL). Unlike most RL applications, in which the decision policy is learned offline on a simulator, we assume no previous knowledge is available to build such a simulator. We thus apply RL in anonlinefashion, i.e., the model and the policy are learned by directly perturbing and monitoring the actual system. Since perturbations generate spurious traffic, we thus need to limit perturbations. This requires learning to be extremely efficient. To this aim, we devise a strategy that learns an approximation of the cost function, while interacting with the system. We then use such an approximation in a Model-Based RL (MB-RL) to speed up convergence. We prove analytically that our strategy brings cache allocation boundedly close to the optimum and stably remains in such an allocation. We show in simulations that such convergence is obtained within few minutes. We also study its fairness, its sensitivity to several scenario characteristics and compare it with a method from the state-of-the-art.

The commercial adoption of Edge Computing (EC) will require pricing schemes that cater to the financial interests of the operators and of the users. Pricing in EC is particularly challenging as it has to take into account the limited amount of edge resources as well as the stochasticity of user workloads due to location-specific workload characteristics and differences in user activity. We formulate the problem of maximizing the revenue of a serverless edge operator through dynamically pricing compute and memory resources under time varying workloads as a sequential decision making problem under uncertainty. We provide analytical results for the optimal pricing strategy in a Markovian setting in steady state. For the general case, we propose a novel Generalized Hidden Parameter Markov Decision Process (GHP-MDP) formulation of the revenue maximization problem, and we propose a dual Bayesian neural network approximator as a solution. The key novelty of the proposed solution is that it can be pre-trained on synthetic traces and adapts fast to previously unseen workload characteristics. We use simulations based on synthetic and real traffic traces to show that the proposed solution is sample-efficient thanks to effective transfer learning, and it outperforms state-of-the-art learning approaches in terms of revenue and learning rate by up to 50% on real traces.

In 5G networks, physical resources can be virtualized and allocated to separate virtual networks (or network slices), with distinct requirements. The Virtual Network Embedding (VNE) problem consists in finding the optimal mapping of virtual resources (virtual links and nodes) onto a physical infrastructure. A recent trend consists in virtualizing 5G networks using Kubernates (K8s), a popular virtualization technology.In this paper we perform an experimental study to show the limit of using the standard K8s deployment strategy when dealing with dynamically arriving slices in a heavy loaded setting. By deploying the virtual components of a slice one by one, standard K8s is prone to wasting resources and energy due to partially deploying slices that, at the end, are found to be infeasible, due to lack of available resources. We propose an alternative K8s deployment strategy that first solves VNE via a Genetic Algorithm and then, for each slice, deploys either all its components or none. Our experimental results show a notable improvement in slice acceptance, energy efficiency and deployment time. Our work shows that it is necessary to adapt cloud native technologies to the specific requirements of telecommunication scenarios, as they are different from the cloud ones for which such technologies were originally developed.

Communication and computation services supporting Connected and Automated Vehicles (CAVs) are characterized by stringent requirements, in terms of response time and relia-bility. Fulfilling these requirements is crucial for ensuring road safety and traffic optimization. The conceptually simple solution of hosting these services in the vehicles increases their cost (mainly due to the installation and maintenance of computation infrastructure) and may drain their battery excessively. Such disadvantages can be tackled via Multi-Access Edge Computing (MEC), consisting in deploying computation capability in network nodes deployed close to the devices (vehicles in this case), such as to satisfy the stringent CAV requirements. However, it is not yet clear under which conditions MEC can support CAV requirements and for which services. To shed light on this question, we conduct a simulation campaign using well-known open-source simulation tools, namely OMNeT++, Simu5G, Veins, INET, and SUMO. We are thus able to provide a reality check on MEC for CAV, pinpointing what are the computation capacities that must be installed in the MEC, to support the different services, and the amount of vehicles that a single MEC node can support. We find that such parameters must vary a lot, depending on the service considered. This study can serve as a preliminary basis for network operators to plan future deployment of MEC to support CAV.

Edge computing (EC) consists of deploying computation resources close to the users, thus enabling low-latency applications, such as augmented reality and online gaming. However, large-scale deployment of edge nodes can be highly impractical and expensive. Besides EC, there is a rising concept known as Vehicular Cloud Computing (VCC). VCC is a computing paradigm that amplifies the capabilities of vehicles by exploiting part of their computational resources, enabling them to participate in services similar to those provided by the EC. The advantage of VCC is that it can opportunistically exploit part of the computation resources already present on vehicles, thus relieving a network operator from the deployment and maintenance cost of EC nodes. However, it is still unknown under which circumstances VCC can enable low-latency applications without EC. In this work, we show that VCC has the potential to effectively supplant EC in urban areas, especially given the higher density of vehicles in such environments. The goal of this paper is to analyze, via simulation, the key parameters determining the conditions under which this substitution of EC by VCC is feasible. In addition, we provide a high level cost analysis to show that VCC is much less costly for a network operator than adopting EC.

An increasing number of applications rely on complex inference tasks that are based on machine learning (ML). Currently, there are two options to run such tasks: either they are served directly by the end device (e.g., smartphones, IoT equipment, smart vehicles), or offloaded to a remote cloud. Both options may be unsatisfactory for many applications: local models may have inadequate accuracy, while the cloud may fail to meet delay constraints. In this paper, we present the novel idea of inference delivery networks (IDNs), networks of computing nodes that coordinate to satisfy ML inference requests achieving the best trade-off between latency and accuracy. IDNs bridge the dichotomy between device and cloud execution by integrating inference delivery at the various tiers of the infrastructure continuum (access, edge, regional data center, cloud). We propose a distributed dynamic policy for ML model allocation in an IDN by which each node dynamically updates its local set of inference models based on requests observed during the recent past plus limited information exchange with its neighboring nodes. Our policy offers strong performance guarantees in an adversarial setting and shows improvements over greedy heuristics with similar complexity in realistic scenarios.

We propose in this paper a coinvestment plan between several stakeholders of different types, namely a physical network owner, operating network nodes, e.g. a network operator or a tower company, and a set of service providers willing to use these resources to provide services as video streaming, augmented reality, autonomous driving assistance, etc. One such scenario is that of deployment of Edge Computing resources. Indeed, although the latter technology is ready, the high Capital Expenditure (CAPEX) cost of such resources is the barrier to its deployment. For this reason, a solid economical framework to guide the investment and the returns of the stakeholders is key to solve this issue. We formalize the coinvestment framework using coalitional game theory. We provide a solution to calculate how to divide the profits and costs among the stakeholders, taking into account their characteristics: traffic load, revenues, utility function. We prove that it is always possible to form the grand coalition composed of all the stakeholders, by showing that our game is convex. We derive the payoff of the stakeholders using the Shapley value concept, and elaborate on some properties of our game. We show our solution in simulation.

Thanks to resource virtualization, Physical Network Operators (PNOs) can share their 5G network to multiple Mobile Virtual Network Operators (MVNOs) which can leverage the shared physical infrastructure to deploy their services up to the edge. This allows much more flexibility with respect to the previous generation of cellular networks: MVNO software components can be placed at different locations, can be allocated a certain amount of virtual resources (e.g., bandwidth, CPU cycles), and be reachable via different paths. To the best of our knowledge, strategies to minimize energy consumption while satisfying Service Level Agreements (SLAs) between the PNO and the MVNOs are still largely missing, particularly if it is required to take the nonlinearity of delays into account. To fill this gap, we formulate the problem of joint placement of software components, routing of user requests and resource dimensioning. SLAs are represented in terms of latency and reliability constraints. Via Column Generation, we obtain exact solutions in real-sized networks. Our numerical results show that we can save up to 50% energy in networks with up to 30 nodes compared to the state-of-the-art algorithms, which are focused on placement or resource minimization.

Edge Computing (EC) allows users to access computing resources at the network frontier, which paves the way for deploying delay-sensitive applications such as Mobile Augmented Reality (MAR). Under the EC paradigm, MAR users connect to the EC server, open sessions and send continuously frames to be processed. The EC server sends back virtual information to enhance the human perception of the world by merging it with the real environment. Resource allocation arises as a critical challenge when several MAR Service Providers (SPs) compete for limited resources at the edge of the network. In this paper, we consider EC in a multi-tenant environment where the resource owner, i.e., the Network Operator (NO), virtualizes the resources and lets SPs run their services using the allocated slice of resources. Indeed, for MAR applications, we focus on two specific resources: CPU and RAM, deployed in some edge node, e.g., a central office. We study the decision of the NO about how to partition these resources among several SPs. We model the arrival and service dynamics of users belonging to different SPs using Erlang queuing model and show that under perfect information, the interaction between the NO and SPs can be formulated as a sub-modular maximization problem under multiple Knapsack constraints. To solve the problem, we use an approximation algorithm, guaranteeing a bounded gap with respect to the optimal theoretical solution. Our numerical results show that the proposed algorithm outperforms baseline proportional allocation in terms of the number of sessions accommodated at the edge for each SP.

We consider in this work Edge Computing (EC) in a multi-tenant environment: the resource owner, i.e., the Network Operator (NO), virtualizes the resources and lets third party Service Providers (SPs - tenants) run their services, which can be diverse and with heterogeneous requirements. Due to confidentiality guarantees, the NO cannot observe the nature of the traffic of SPs, which is encrypted. This makes resource allocation decisions challenging, since they must be taken based solely on observed monitoring information.We focus on one specific resource, i.e., cache space, deployed in some edge node, e.g., a base station. We study the decision of the NO about how to partition cache among several SPs in order to minimize the upstream traffic. Our goal is to optimize cache allocation using purely data-driven, model-free Reinforcement Learning (RL). Differently from most applications of RL, in which the decision policy is learned offline on a simulator, we assume no previous knowledge is available to build such a simulator. We thus apply RL in an online fashion, i.e., the policy is learned by directly perturbing the actual system and monitoring how its performance changes. Since perturbations generate spurious traffic, we also limit them. We show in simulation that our method rapidly converges toward the theoretical optimum, we study its fairness, its sensitivity to several scenario characteristics and compare it with a method from the state-of-the-art. Our code to reproduce the results is available as open source.1

Within network slicing, Virtual Network Embedding has been vastly studied, i.e., deciding in which physical nodes and links to place virtual functions and links. However, the performance of slices does not only depend on where virtual functions and links are placed, but also on how much resources they can use, which has been mostly neglected in the literature. We thus propose a method for optimal resource dimensioning, via dimensioning capacities of multiple Jackson networks (one per slice) co-existing in the same resource-constrained network. Despite the long history of Jackson networks, we are the first, to the best of our knowledge, to model such a problem. The objective is to minimize energy consumption while satisfying the latency requirements of heterogeneous service providers. We show numerically that our solution is able to achieve both goals, differently from classic approaches, which assume that the amount of resources assigned to slices is fixed a-priori.

Network slicing in the upcoming Telecom generation is a fundamental feature which is deployed to satisfy the various demands in term of data rate and latency. On the other hand, it is seen as a topic that imposes other questions such as the coexistence of physical and virtual functions. In this context, we consider the resource management problem for 5G networks slicing since the solution searches to optimally allocate multiple Virtual Network Requests (VNRs) on a substrate virtualized physical network. In this demo, we present an integrated framework that uses an agile service platform (Kube5G) to deploy one of the VNE proposed solutions with zero-touch configuration. The aim of this integration is to validate the proposed solution and to practically study the performance differences among multiple algorithms that will be conducted later as well. The overview of the process is shown in steps as exposing the resources’ availability of the Physical Nodes (PN), which will be the input of the orchestration algorithm. Successively, the last takes the suitable decision to deploy VNRs on a substrate network, based on the VNRs’ demands such as CPU and radio resources and PNs’ availability. Afterwards, the decision will be sent to the platform to host the virtual nodes on the chosen physical machines. Bearing in mind that the essential objective of this algorithm is to achieve a better resource usage and increase the VNR acceptance ratio on the physical nodes with respect to the constraints that might affect the performance.

We address the problem of a Network Operator (NO) owning limited resources at the network edge. The NO wishes to enable advanced services, by virtualizing and allocating such resources among multiple tenants, i.e., third-party Service Providers, co-existing at the edge, with different Quality of Service (QoS) constraints. The NO applies a resource allocation policy with the objective of minimizing energy consumption via switching off non-used resources while guaranteeing tenants QoS requirements. We propose a resource allocation policy based on Markov Decision Processes (MDP). In simulation we show that our policy is able to reduce energy consumption, by turning off unused resources, while meeting heterogeneous SP requirements. Our code is available as open source.

Edge Computing exploits computational capabilities deployed at the very edge of the network to support applications with low latency requirements. Such capabilities can reside in small embedded devices that integrate dedicated hardware - e.g., a GPU - in a low cost package. But these devices have limited computing capabilities compared to standard server grade equipment. When deploying an Edge Computing based application, understanding whether the available hardware can meet target requirements is key in meeting the expected performance. In this paper, we study the feasibility of deploying Augmented Reality applications using Embedded Edge Devices (EEDs). We compare such deployment approach to one exploiting a standard dedicated server grade machine. Starting from an empirical evaluation of the capabilities of these devices, we propose a simple theoretical model to compare the performance of the two approaches. We then validate such model with NS-3 simulations and study their feasibility. Our results show that there is no one-fits-all solution. If we need to deploy high responsiveness applications, we need a centralized server grade architecture and we can in any case only support very few users. The centralized architecture fails to serve a larger number of users, even when low to mid responsiveness is required. In this case, we need to resort instead to a distributed deployment based on EEDs.

The 5G telecommunication ecosystem is expected to dynamically support new and various applications from the industrial and the service sectors that are very heterogeneous in terms of QoS and resources’ requirements. In this context, a promising important concept for network resource management is emerging, denoted by Network Slicing. It involves decisions on embedding and managing several virtual networks on the same physical resources. This problem in its simplified form can be modeled by the Virtual Network Embedding (VNE) problem. In this paper, we propose a new resolution method, in which we first reduce the set of admitted routes and then solve an integer program. Our proposed approach is then compared to the optimal solution and to a method from the state of the art. Obtained results show that our approach provides good result in terms of slice acceptance ratio and resource consumption while reducing the overall complexity and runtime.

In this paper, we consider the Virtual Network Embedding (VNE) problem for 5G networks slicing. This consists in optimally allocating multiple Virtual Networks (VN) on a substrate virtualized physical network while maximizing among others, resource utilization, maximum number of placed VNs and network operator's benefit. We solve the online version of the problem where slices arrive over time. We propose the use of the Nested Rollout Policy Adaptation (NRPA) algorithm, a variant of the well known Monte Carlo Tree Search (MCTS). Both algorithms learn by randomly simulating the embedding, but NRPA also learns how to perform better simulations over time. Performance analysis with different scenarios, show that NRPA improves acceptance and reward ratios (by up to 69% and 65%). We also show how a smart initialization of the learning process can help improve the results furthermore (up to a 12.5% increase of acceptance ratio).

Under the paradigm of Edge Computing (EC), a Network Operator (NO) deploys computational resources at the network edge and let third-party Service Providers (SPs) run on top of them, as tenants. Besides the clear advantages for SPs and final users thanks to the vicinity of computation nodes, a NO aims to allocate edge resources in order to increase its own utility, including bandwidth saving, operational cost reduction, QoE for its users, etc. However, while the number of third-party services competing for edge resources is expected to dramatically grow, the resources deployed cannot increase accordingly, due to physical limitations. Therefore, smart strategies are needed to fully exploit the potential of EC, despite its constrains. To this aim, we propose to leverage service adaptability, a dimension that has mainly been neglected so far: each service can adapt to the amount of resources that the NO has allocated to it, balancing the fraction of service computation performed at the edge and relying on remote servers, e.g., in the Cloud, for the rest. We propose EdgeMORE, a resource allocation strategy in which SPs express their capabilities to adapt to different resource constraints, by declaring the different configurations under which they are able to run, specifying the resources needed and the utility provided to the NO. The NO then chooses the most convenient option per each SP, in order to maximize the total utility. We formalize EdgeMORE as a Integer Linear Program. We show via simulation that EdgeMORE greatly improves EC utility with respect to the standard where no multiple options for running services are allowed.

Mobility on Demand (MoD) services are revolutionizing the way people move in cities around the world and are often considered a convenient alternative to public transit. MoD is usually intended as a door-to-door service. However, there has been recent interest toward consolidating, e.g., aggregating, the travel demand by limiting the number of admitted stop locations. This implies users have to walk from/to their intended origin/destination.The contribution of this paper is a systematic study the impact of consolidation on the operator cost and on user QoS. We introduce a MoD system where pick-ups and drop-offs can only occur in a limited subset of admitted stop locations. The density of such locations is a system parameter: the less the density, the more the user demand is consolidated. We show that, by decreasing stop density, we can increase system capacity (number of passengers we are able to serve). On the contrary, increasing it, we can improve QoS. The system is tested in AMoDSim, an open-source simulator.This work is a first step toward flexible mobility services that are able to autonomously re-conFigure themselves, favoring capacity or QoS, depending on the amount of travel demand coming from users. In other words, the services we envisage in this work shift their operational mode to any intermediate point in the range from a taxi-like door-to-door service to a bus-like service, with few served stops and more passengers on-board.

In-network caching is an appealing solution to cope with the increasing bandwidth demand of video, audio, and data transfer over the Internet. Nonetheless, in order to protect consumer privacy and their own business, content providers (CPs) increasingly deliver encrypted content, thereby preventing Internet service providers (ISPs) from employing traditional caching strategies, which require the knowledge of the objects being transmitted. To overcome this emerging tussle between security and efficiency, in this paper we propose an architecture in which the ISP partitions the cache space into slices, assigns each slice to a different CP, and lets the CPs remotely manage their slices. This architecture enables transparent caching of encrypted content and can be deployed in the very edge of the ISP's network (i.e., base stations and femtocells), while allowing CPs to maintain exclusive control over their content. We propose an algorithm, called SDCP, for partitioning the cache storage into slices so as to maximize the bandwidth savings provided by the cache. A distinctive feature of our algorithm is that ISPs only need to measure the aggregated miss rates of each CP, but they need not know the individual objects that are requested. We prove that the SDCP algorithm converges to a partitioning that is close to the optimal, and we bound its optimality gap. We use simulations to evaluate SDCP's convergence rate under stationary and nonstationary content popularity. Finally, we show that SDCP significantly outperforms traditional reactive caching techniques, considering both CPs with perfect and with imperfect knowledge of their content popularity.

Albeit an important goal of caching is traffic reduction, a perhaps even more important aspect follows from the above achievement: the reduction of internet service provider (ISP) operational costs that comes as a consequence of the reduced load on transit and provider links. Surprisingly, to date this crucial aspect has not been properly taken into account in cache design. In this paper, we show that the classic caching efficiency indicator, i.e., the hit ratio, conflicts with cost. We therefore propose a mechanism whose goal is the reduction of cost and, in particular, we design a cost-aware (CoA) cache decision policy that, leveraging price heterogeneity among external links, tends to store with more probability the objects that the ISP has to retrieve through the most expensive links. We provide a model of our mechanism, based on Che's approximation, and, by means of a thorough simulation campaign, we contrast it with traditional cost-blind schemes, showing that CoA yields a significant cost saving, that is furthermore consistent over a wide range of scenarios. We show that CoA is easy to implement and robust, making the proposal of practical relevance.

Embedded systems design requires conflicting objectives to be optimized with an appropriate choice of hardware-software parameters. A simulation campaign can guide the design in finding the best trade-offs, but due to the big number of possible configurations, it is often unfeasible to simulate them all. For these reasons, design space exploration algorithms aim at finding near-optimal system configurations by simulating only a subset of them. In this work, we present PS, a new multiobjective optimization algorithm, and evaluate it in the context of the embedded system design. The basic idea is to recognize interesting regions—that is, regions of the configuration space that provide better configurations with respect to other ones. PS evaluates more configurations in the interesting regions while less thoroughly exploring the rest of the configuration space. After a detailed formal description of the algorithm and the underlying concepts, we show a case study involving the hardware/software exploration of a VLIW architecture. Qualitative and quantitative comparisons of PS against a well-known multiobjective genetic approach demonstrate that while not outperforming it in terms of Pareto dominance, the proposed approach can balance the uniformity and granularity qualities of the solutions found, obtaining more extended Pareto fronts that provide a wider view of the potentiality of the designed device. Therefore, PS represents a further valid choice for the designer when objective constrains allow it.

Caching is frequently used by Internet Service Providers as a viable technique to reduce the latency perceived by end users, while jointly offloading network traffic. While the cache hit-ratio is generally considered in the literature as the dominant performance metric for such type of systems, in this paper we argue that a critical missing piece has so far been neglected. Adopting a radically different perspective, in this paper we explicitly account for the cost of content retrieval, i.e. the cost associated to the external bandwidth needed by an ISP to retrieve the contents requested by its customers. Interestingly, we discover that classical cache provisioning techniques that maximize cache efficiency (i.e., the hit-ratio), lead to suboptimal solutions with higher overall cost. To show this mismatch, we propose two optimization models that either minimize the overall costs or maximize the hit-ratio, jointly providing cache sizing, object placement and path selection. We formulate a polynomial-time greedy algorithm to solve the two problems and analytically prove its optimality. We provide numerical results and show that significant cost savings are attainable via a cost-aware design.

We propose a methodology to gauge the extent of queueing delay (aka bufferbloat) in the Internet, based on purely passive measurement of TCP traffic. We implement our methodology in Tstat and make it available as open source software. We leverage Deep Packet Inspection (DPI) and behavioral classification of Tstat to breakdown the queueing delay across different applications, in order to evaluate the impact of bufferbloat on user experience. We show that there is no correlation between the ISP traffic load and the queueing delay, thus confirming that bufferbloat is related only to the traffic of each single user (or household). Finally, we use frequent itemset mining techniques to associate the amount of queueing delay seen by each host with the set of its active applications, with the goal of investigating the root cause of bufferbloat.