Search

International audience; EdgeNet is a public Kubernetes cluster dedicated to network and distributed systems research, supporting experiments that are deployed concurrently by independent groups. Its nodes are hosted by multiple institutions around the world. It represents a departure from the classic Kubernetes model, where the nodes that are available to a single tenant reside in a small number of well-interconnected data centers. The free open-source EdgeNet code extends Kubernetes to the edge, making three key contributions: multi-tenancy, geographical deployments, and single-command node installation. We show that establishing a public Kubernetes cluster over the internet, with multiple tenants and multiple hosting providers is viable. Preliminary results also indicate that the EdgeNet testbed that we run provides a satisfactory environment to run a variety of experiments with minimal network overhead.

Presently, the cloud computing environment attracts many application developers to deploy their web applications on cloud data centers. Kubernetes, a well-known container orchestration for deploying web applications on cloud systems, offers an automatic scaling feature to meet clients’ ever-changing demands with the reactive approach. This paper proposes a system architecture based on Kubernetes with a proactive custom autoscaler using a deep neural network model to handle the workload during run time dynamically. The proposed system architecture is designed based on the Monitor–Analyze–Plan–Execute (MAPE) loop. The main contribution of this paper is the proactive custom autoscaler, which focuses on the analysis and planning phases. In analysis phase, Bidirectional Long Short-term Memory (Bi-LSTM) is applied to predict the number of HTTP workloads in the future. In the planning phase, a cooling-down time period is implemented to mitigate the oscillation problem. In addition, a resource removal strategy is proposed to remove a part of the resources when the workload decreases, so that the autoscaler can handle it faster when the burst of workload happens. Through experiments with two different realistic workloads, the Bi-LSTM model achieves better accuracy not only than the Long Short-Term Memory model but also than the state-of-the-art statistical auto-regression integrated moving average model in terms of short- and long-term forecasting. Moreover, it offers 530 to 600 times faster prediction speed than ARIMA models with different workloads. Furthermore, as compared to the LSTM model, the Bi-LSTM model performs better in terms of resource provision accuracy and elastic speedup. Finally, it is shown that the proposed proactive custom autoscaler outperforms the default horizontal pod autoscaler (HPA) of the Kubernetes in terms of accuracy and speed when provisioning and de-provisioning resources.

Microservices, Continuous Integration and Delivery, Docker, DevOps, Infrastructure as Code—these are the current trends and buzzwords in the technological world of 2020. A popular tool which can facilitate the deployment and maintenance of microservices is Kubernetes. Kubernetes is a platform for running containerized applications, for example microservices. There are two main questions which answer was important for us: how to deploy Kubernetes itself and how to ensure that the deployment fulfils the needs of a production environment. Our research concentrates on the analysis and evaluation of Kubernetes cluster as the software production environment. However, firstly it is necessary to determine and evaluate the requirements of production environment. The paper presents the determination and analysis of such requirements and their evaluation in the case of Kubernetes cluster. Next, the paper compares two methods of deploying a Kubernetes cluster: kops and eksctl. Both of the methods concern the AWS cloud, which was chosen mainly because of its wide popularity and the range of provided services. Besides the two chosen methods of deployment, there are many more, including the DIY method and deploying on-premises.

[EN] This paper describes an approach to integrate the jobs management of High Performance Computing (HPC) infrastructures in cloud architectures by managing HPC workloads seamlessly from the cloud job scheduler. The paper presents hpc-connector, an open source tool that is designed for managing the full life cycle of jobs in the HPC infrastructure from the cloud job scheduler interacting with the workload manager of the HPC system. The key point is that, thanks to running hpc-connector in the cloud infrastructure, it is possible to reflect in the cloud infrastructure, the execution of a job running in the HPC infrastructure managed by hpc-connector. If the user cancels the cloud-job, as hpc-connector catches Operating System (OS) signals (for example, SIGINT), it will cancel the job in the HPC infrastructure too. Furthermore, it can retrieve logs if requested. Therefore, by using hpc-connector, the cloud job scheduler can manage the jobs in the HPC infrastructure without requiring any special privilege, as it does not need changes on the Job scheduler. Finally, we perform an experiment training a neural network for automated segmentation of Neuroblastoma tumours in the Prometheus supercomputer using hpc-connector as a batch job from a Kubernetes infrastructure., The work presented in this article has been partially funded by the regional government of the Comunitat Valenciana (Spain), co-funded by the European Union ERDF funds (European Regional Development Fund) of the Comunitat Valenciana 2014¿2020, with reference IDIFEDER/2018/032 (High-Performance Algorithms for the Modeling, Simulation and early Detection of diseases in Personalized Medicine). The work is also co-funded by PRIMAGE (PRedictive In-silico Multiscale Analytics to support cancer personalised diaGnosis and prognosis, empowered by imaging biomarkers) a Horizon 2020 RIA project funded under the topic SC1-DTH-07-2018 by the European Commission, with grant agreement no: 826494.

Distributed cloud storage solutions are currently gaining high momentum in industry and academia. The enterprise data volume growth and the recent tendency to move as much as possible data to the cloud is strongly stimulating the storage market growth. In this context, and as a main requirement for cloud native applications, it is of utmost importance to guarantee resilience of the deployed applications and the infrastructure. Indeed, with failures frequently occurring, a storage system should quickly recover to guarantee service availability. In this paper, we focus on containerized cloud storage, proposing a resilience monitoring solution for the recently developed Rook storage operator. While, Rook brings storage systems into a cloud-native container platform, in this paper we design an additional module to monitor and evaluate the resilience of the Rook-based system. Our proposed module is validated in a production environment, with software components generating a constant load and a controlled removal of system elements to evaluate the self-healing capability of the storage system. Failure recovery time revealed to be 41 and 142 seconds on average for a 32GB and a 215GB object storage device respectively.

Service meshes can be seen as an infrastructure layer for microservice-based applications that are specifically suited for distributed application architectures. It is the goal to introduce the concept of service meshes and its use for microservices with the example of an open source service mesh called Istio. This paper gives an introduction into the service mesh concept and its relation to microservices. It also gives an overview of selected features provided by Istio as relevant to the above concept and provides a small sample setup that demonstrates the core features.

This paper proposes an energy management system (EMS) architecture based on the Kubernetes container cluster to solve the problem of traditional EMSs being unable to simultaneously achieve high reliability and high resource utilization. Container cluster technology is used to encapsulate, isolate and deploy applications, which solves the problem of low system reliability caused by interlocking failures. Discrete Markov theory is applied to propose a dynamic Pod fault-tolerant EMS model. The results of the solution model are used to adjust the Pod redundancy in real-time to achieve the highest reliability to satisfy physical resource constraints. The results of the performance analysis show that the reliability of the proposed architecture is 99.9999504%. Compared with the EMS of the service-oriented architecture (SOA), the annual failure time is reduced from 3.83 minutes to 0.26 minutes. The comprehensive utilization of hardware resources increases by approximately 20%, and performance indicators such as the peak access success rate improve significantly. The proposed architecture is implemented in a real-power system, with good operating results and broad application prospects.

This paper is an extension of work originally presented in SoftCOM 2019 [1]. The novelty of this work reside in its focused improvement of our scheduling algorithm towards its usage on a real 5G infrastructure. Industrial IoT applications are often designed to run in a distributed way on the devices and controller computers with strict service requirements for the nodes and the links between them. 5G, especially in concomitance with Edge Computing, will provide the desired level of connectivity for these setups and it will permit to host application run-time components in edge clouds. However, allocation of the edge cloud resources for Industrial IoT (IIoT) applications, is still commonly solved by rudimentary scheduling techniques (i.e. simple strategies based on CPU usage and device readiness, employing very few dynamic information). Orchestrators inherited from the cloud computing, like Kubernetes, are not satisfying to the requirements of the aforementioned applications and are not optimized for the diversity of devices which are often also limited in capacity. This design is especially slow in reacting to the environmental changes. In such circumstances, in order to provide a proper solution using these tools, we propose to take the physical, operational and network parameters (thus the full context of the IIoT application) into consideration, along with the software states and orchestrate the applications dynamically.

Modern cloud orchestrators like Kubernetes provide a versatile and robust way to host applications at scale. One of their key features is autoscaling, which automatically adjusts cloud resources (compute, memory, storage) in order to adapt to the demands of applications. However, the scope of cloud autoscaling is limited to the datacenter hosting the cloud and it doesn't apply uniformly to the allocation of network resources. In I/O-constrained or data-in-motion use cases this can lead to severe performance degradation for the application. For example, when the load on a cloud service increases and the Wide Area Network (WAN) connecting the datacenter to the Internet becomes saturated, the application flows experience an increase in delay and loss. In many cases this is dealt with overprovisioning network capacity, which introduces additional costs and inefficiencies. On the other hand, thanks to the concept of "Network as Code", the WAN exposes a set of APIs that can be used to dynamically allocate and de-allocate capacity on-demand. In this paper we propose extending the concept of cloud autoscaling into the network to address this limitation. This way, applications running in the cloud can communicate their networking requirements, like bandwidth or traffic profile, to a Software-Defined Networking (SDN) controller or Network as a Service (NaaS) platform. Moreover, we aim to define the concepts of vertical and horizontal autoscaling applied to networking. We present a prototype that automatically allocates bandwidth to the underlay network, according to the requirements of the applications hosted in Kubernetes. Finally, we discuss open research challenges., 6 pages, 5 figures

International audience; Internet application users are increasingly concerned about the way applications handle their personal data. However, manually checking whether applications actually respect the claims made in their privacy policy is both error-prone and time-consuming. This paper claims that the privacy compliance of applications hosted in cloud or fog computing platforms can and should be automatically carried by the platform itself. We discuss the feasibility of unintrusive and application-agnostic monitoring in the platform layer to check the privacy compliance of applications. First, the platform may monitor an application's privacy-oriented behavior through signals such as its network traffic characteristics. Second, these signals can be analyzed and compared with the principles found in the application's privacy policy. We present a procedure based on machinelearning techniques to identify the type of data being shared by applications with external third-parties even if the application uses encrypted communications. Our classifiers identify traffic samples of applications with 86% accuracy.

With the rapidly growing popularity of cloud services, the cloud computing faces critical challenges to orchestrate the deployment and operation of cloud applications on heterogenous cloud platforms. Cloud applications are built on a platform model that abstracts away underlying platform-specific details, so that their orchestration can benefit from the abstract view and flexibility of the underlying platform configuration. However, considerable efforts are still required to properly manage complicated cloud applications. This paper proposes a model-driven approach to cloud application orchestration which promotes the concerns of distinct roles for cloud system provisioning and operation. By establishing a set of capabilities as modeling constructs, our approach allows TOSCA-based application topology itself and its orchestration needs to be specified in a way to provide a more targeted support for different needs and concerns of application developers and operators. With novel orchestration features like application topology description, platform capability modeling, and role-awareness for cloud application orchestration, it can significantly reduce the complexity of application orchestration in diverse cloud environments. To show the feasibility and effectiveness of our proposal for cloud application orchestration, we present a proof-of-concept orchestration system implementation and evaluate its deployment and orchestration results in a Kubernetes cluster.

State-of-the-art scenarios, such as Internet of Things (IoT) and Smart Cities, have recently arisen. They involve the processing of huge data sets under strict time requirements, rendering the use of cloud resources unfeasible. For this reason, Fog computing has been proposed as a solution, however, there remains a need for intelligent allocation decisions, in order to make it a fully usable solution in such contexts. In this paper, a network-aware scheduling algorithm is presented, which aims to select the fog node most suitable for the execution of an application within a given deadline. This decision is made taking the status of the network into account. This scheduling algorithm was implemented as an extension to the Kubernetes default scheduler, and compared with existing proposals in the literature. The comparison shows that our proposal is the only one that can execute all the submitted jobs within their deadlines (i.e., no job is rejected or executed exceeding its deadline) with certain configurations in some of the scenarios tested, thus obtaining an optimal solution in such scenarios.

Global sensitivity analysis, like variance-based methods for massive raster datasets, is especially computationally costly and memory-intensive, limiting its applicability for commodity cluster computing. The computational effort depends mainly on the number of model runs, the spatial, spectral, and temporal resolutions, the number of criterion maps, and the model complexity. The current Spatially-Explicit Uncertainty and Sensitivity Analysis (SEUSA) approach employs a cluster-based parallel and distributed Python–Dask solution for large-scale spatial problems, which validates and quantifies the robustness of spatial model solutions. This paper presents the design of a framework to perform SEUSA as a Service in a cloud-based environment scalable to very large raster datasets and applicable to various domains, such as landscape assessment, site selection, risk assessment, and land-use management. It incorporates an automated Kubernetes service for container virtualization, comprising a set of microservices to perform SEUSA as a Service. Implementing the proposed framework will contribute to a more robust assessment of spatial multi-criteria decision-making applications, facilitating a broader access to SEUSA by the research community and, consequently, leading to higher quality decision analysis.

Serverless computing is a cloud-based execution paradigm that allows provisioning resources on-demand, freeing developers from infrastructure management and operational concerns. It typically involves deploying workloads as stateless functions that take no resources when not in use, and is meant to scale transparently. To make serverless effective, providers impose limits on a per-function level, such as maximum duration, fixed amount of memory, and no persistent local storage. These constraints make it challenging for data-intensive workloads to take advantage of serverless because they lead to sharing significant amounts of data through remote storage. In this paper, we build a performance model for serverless workloads that considers how data is shared between functions, including the amount of data and the underlying technology that is being used. The model's accuracy is assessed by running a real workload in a cluster using Knative, a state-of-the-art serverless environment, showing a relative error of 5.52%. With the proposed model, we evaluate the performance of data-intensive workloads in serverless, analyzing parallelism, scalability, resource requirements, and scheduling policies. We also explore possible solutions for the data-sharing problem, like using local memory and storage. Our results show that the performance of data-intensive workloads in serverless can be up to 4.32= faster depending on how these are deployed. This work was partially supported by the Ministry of Economy of Spain under contract TIN2015-65316-P, the Ministry of Science under contract PID2019-107255GB-C21/AEI/10.13039/501100011033, and the Generalitat de Catalunya under contract 2014SGR1051.

International audience; Microservices represent a popular paradigm to construct large-scale applications in many domains thanks to benefits such as scalability, flexibility, and agility. However, it is difficult to manage and operate a microservice system due to its high dynamics and complexity. In particular, the frequent updates of microservices lead to the absence of historical failure data, where the current automatic recovery methods fail short. In this paper, we propose an automatic recovery method named MicroRAS, which requires no historical failure data, to mitigate performance issues in microservice systems. MicroRAS is a model-driven method that selects the appropriate recovery action with a trade-off between the effectiveness and recovery time of actions. It estimates the effectiveness of an action in terms of its effects of recovering the pinpointed faulty service and its effects of interfering with other services. The estimation of action effects is based on a system-state model represented by an attributed graph that tracks the propagation of effects. For the experimental evaluation, several types of anomalies are injected into a microservice system based on Kubernetes, which also serves a real-world workload. The corresponding benchmarks show that the actions selected by MicroRAS can recover the faulty services by 94.7%, and reduce the interference to other services by at least 44.3% compared to baseline methods.

International audience; With the proliferation of use-cases envisioned to be supported by 5G networks, the focus is not only on the performance, but also on the service agility and elasticity. Cloud native principal is a methodology of designing lightweight, isolatedcontext, and deployable at scale applications that natively exploit the features of cloud. However, supporting telco applications (e.g., 4G/5G services) in the cloud brings up many challenges, such as the coexistence of physical and virtual functions, (near) realtime resource provisioning, service continuity, and strict latency and data rates. In this paper, we propose Kube5G, as a realizable agile service platform in support of 4G/5G cloud native applications. Kube5G introduces a novel approach in building and packaging a cloud-native compliant telco network function (NF) in a form of nested well-defined layers. In addition, we present a workflow for continuous integration and development operations in support of multi-version network functions (physical, virtual and cloud functions). We also present a concrete prototype implementation of Kube5G with experimental results indicating its efficiency and highlighting user and network perceived performance.

Kubernetes, an open-source container orchestration platform, enables high availability and scalability through diverse autoscaling mechanisms such as Horizontal Pod Autoscaler (HPA), Vertical Pod Autoscaler and Cluster Autoscaler. Amongst them, HPA helps provide seamless service by dynamically scaling up and down the number of resource units, called pods, without having to restart the whole system. Kubernetes monitors default Resource Metrics including CPU and memory usage of host machines and their pods. On the other hand, Custom Metrics, provided by external software such as Prometheus, are customizable to monitor a wide collection of metrics. In this paper, we investigate HPA through diverse experiments to provide critical knowledge on its operational behaviors. We also discuss the essential difference between Kubernetes Resource Metrics (KRM) and Prometheus Custom Metrics (PCM) and how they affect HPA&rsquo, s performance. Lastly, we provide deeper insights and lessons on how to optimize the performance of HPA for researchers, developers, and system administrators working with Kubernetes in the future.

Cloud applications are exposed to workloads whose intensity can change unpredictably over time. Hence, the ability to quickly scale the amount of computing resources provisioned to applications is essential to minimize costs while providing reliable services. In this context,containers are deemed to be a promising technology to enable fast elasticity in resource allocation schemes.In this paper, we propose and experimentally test an efficient container-based cloud computing provisioning system. First, we address the container deployment problem and discuss how to manage container provisioning and scaling. Second, we devise are source management mechanism leveraging on both admission control and auto-scaling techniques. We propose to drive auto-scaling decisions through a Q-Learning algorithm, which is agnostic to the specific computing environment, and proceeds based only on the load of the physical processors assigned to a container. We evaluate our solution in two experimental setups,and show that it yields significant advantages when compared to popular container managers such as Kubernetes. TRUE pub

Machine Learning (ML) and Artificial Intelligence (AI) have a dependency on data sources to train, improve and make predictions through their algorithms. With the digital revolution and current paradigms like the Internet of Things, this information is turning from static data into continuous data streams. However, most of the ML/AI frameworks used nowadays are not fully prepared for this revolution. In this paper, we proposed Kafka-ML, an open-source framework that enables the management of TensorFlow ML/AI pipelines through data streams (Apache Kafka). Kafka-ML provides an accessible and user-friendly Web User Interface where users can easily define ML models, to then train, evaluate and deploy them for inference. Kafka-ML itself and its deployed components are fully managed through containerization technologies, which ensure its portability and easy distribution and other features such as fault-tolerance and high availability. Finally, a novel approach has been introduced to manage and reuse data streams, which may lead to the (no) utilization of data storage and file systems., 10 pages, 10 figures

The OASIS standard TOSCA provides a portable means for specifying multi-service applications and automating their deployment. Despite TOSCA is widely used in research, it is currently not supported by the production-ready deployment technologies daily used by practitioners, hence resulting in a gap between the state-of-the-art in research and the state-of-practice in industry. To help bridging this gap, we identified TOSCA Light, a subset of TOSCA enabling the transformation of compliant deployment models to the vast majority of deployment technology-specific models used by practitioners nowadays. In this paper, we demonstrate TOSCA Lightning by two contributions. We (i) present an integrated toolchain for specifying multi-service applications with TOSCA Light and transforming them into different production-ready deployment technologies. Additionally, we (ii) demonstrate the toolchain’s effectiveness based on a third-party application and Kubernetes.

The growing adoption of software architectures based on microservices, and consequently the proliferation of the use of containers in environments with hybrid infrastructures (local data centers and public cloud providers), poses several challenges in terms of integrating this new paradigm into already existing solutions. In this context, this article initially focuses on the challenges of integrating a container orchestration platform (Kubernetes) with existing solutions in the data center, namely: Network infrastructure (Cisco ACI - Application Centric Infrastructure), load balancing (F5) and storage (NFS). In a second stage, this paper also addresses the issue of integration with orchestration platforms hosted in a public cloud (Microsoft Azure). This integration will allow the movement of containers from the local data center to the cloud without having to reconFigure the network infrastructure or the current security policy., A crescente adoção de arquiteturas de software baseadas em microserviços, e consequentemente a proliferação da utilização de containers em ambientes com infraestruturas híbridas (centros de dados locais e operadores de clouds públicas), lança vários desafios do ponto de vista da integração deste novo paradigma nas soluções existentes. Neste contexto, este artigo centra-se inicialmente nos desafios de integrar uma plataforma de orquestração de containers (Kubernetes) com soluções já existentes no centro de dados, nomeadamente com a componente de rede (Cisco ACI – Application Centric Infrastructure), com o balanceador de carga (F5) e armazenamento (NFS). Num segundo momento é também abordada a integração com plataformas de orquestração alojadas numa cloud pública (Microsoft Azure). Esta integração vai permitir a movimentação de containers do centro de dados local para a cloud sem necessitar de reconfigurar a infraestrutura de rede ou as políticas de segurança em vigor., This work has been supported by FCT – Fundação para a Ciência e Tecnologia within the R&D Units Project Scope: UIDB/00319/2020

The Internet of Things (IoT) is a network widely used with the purpose of connecting almost everything, everywhere to the Internet. To cope with this goal, low cost nodes are being used, otherwise, it would be very expensive to expand so fast. These networks are set up with small distributed devices (nodes) that have a power supply, processing unit, memory, sensors, and wireless communications. In the market, we can find different alternatives for these devices, such as small board computers (SBCs), e.g., Raspberry Pi (RPi)), with different features. Usually these devices run a coarse version of a Linux operating system. Nevertheless, there are many scenarios that require enhanced computational power that these nodes alone are unable to provide. In this context, we need to introduce a kind of collaboration among the devices to overcome their constraints. We based our solution in a combination of clustering techniques (building a mesh network using their wireless capabilities), at the same time we try to orchestrate the resources in order to improve their processing capabilities in an elastic computing fashion. This paradigm is called fog computing on IoT. We propose in this paper the use of cloud computing technologies, such as Linux containers, based on Docker, and a container orchestration platform (COP) to run on the top of a cluster of these nodes, but adapted to the fog computing paradigm. Notice that these technologies are open source and developed for Linux operating system. As an example, in our results we show an IoT application for soundscape monitoring as a proof of concept that it will allow us to compare different alternatives in its design and implementation, in particular, with regard to the COP selection, between Docker Swarm and Kubernetes. We conclude that using and combining these techniques, we can improve the overall computation capabilities of these IoT nodes within a fog computing paradigm.

Managing the large volumes of IoT and M2M traffic requires the evaluation of the scalability and reliability for all the components in the end-to-end system. This includes connectivity, mobile network functions, and application or services receiving and processing the data from end devices. Firstly, this paper discusses the design of a containerized IoT and M2M application and the mechanisms for delivering automated scalability and high availability when deploying it in: (1) the edge using balenaCloud, (2) the Amazon Web Services cloud with EC2 instances, and (3) the dedicated Amazon Web Services IoT service. The experiments showed that there are no significant differences between edge and cloud deployments regarding resource consumption. Secondly, the solutions for scaling the 4G/5G network functions and mobile backhaul that provide the connectivity between devices and IoT/M2M applications are analyzed. In this case, the scalability and high availability of the 4G/5G components are provided by Kubernetes. The experiments showed that our proposed scaling algorithm for network slicing managed with SDN guarantees the necessary radio and network resources for end-to-end high availability.

How to design efficient scheduling strategy for different environments is a hot topic in cloud computing. In the private cloud of computer science labs in universities, there are several kinds of tasks with different resource requirements, constraints, and lifecycles such as IT infrastructure tasks, course design tasks submitted by undergraduate students, deep learning tasks and and so forth. Taking the actual needs of our laboratory as an instance, these tasks are analyzed, and scheduled respectively by different scheduling strategies. The Batch Scheduler is designed to process tasks in rush time to improve system throughput. Dynamic scheduling algorithm is proposed to tackle long-term lifecycle tasks such as deep learning tasks which are hungry for GPU resources and have dynamically changing priorities. Experiments show that the scheduling strategies proposed in this paper improve resource utilization and efficiency.

Geospatial big data analysis (GBDA) is extremely significant for time-constraint applications such as disaster response. However, the time-constraint analysis is not yet a trivial task in the cloud computing environment. Spatial query processing (SQP) is typical computation-intensive and indispensable for GBDA, and the spatial range query, join query, and the nearest neighbor query algorithms are not scalable without using MapReduce-liked frameworks. Parallel SQP algorithms (PSQPAs) are trapped in screw-processing, which is a known issue in Geoscience. To satisfy time-constrained GBDA, we propose an elastic SQP approach in this paper. First, Spark is used to implement PSQPAs. Second, Kubernetes-managed Core Operation System (CoreOS) clusters provide self-healing Docker containers for running Spark clusters in the cloud. Spark-based PSQPAs are submitted to Docker containers, where Spark master instances reside. Finally, the horizontal pod auto-scaler (HPA) would scale-out and scale-in Docker containers for supporting on-demand computing resources. Combined with an auto-scaling group of virtual instances, HPA helps to find each of the five nearest neighbors for 46,139,532 query objects from 834,158 spatial data objects in less than 300 s. The experiments conducted on an OpenStack cloud demonstrate that auto-scaling containers can satisfy time-constraint GBDA in clouds.

In this paper we present Foggy, an architectural framework and software platform based on Open Source technologies. Foggy orchestrates application workload, negotiates resources and supports IoT operations for multi-tier, distributed, heterogeneous and decentralized Cloud Computing systems. Foggy is tailored for emerging domains such as 5G Networks and IoT, which demand resources and services to be distributed and located close to data sources and users following the Fog Computing paradigm. Foggy provides a platform for infrastructure owners and tenants (i.e., application providers) offering functionality of negotiation, scheduling and workload placement taking into account traditional requirements (e.g. based on RAM, CPU, disk) and non-traditional ones (e.g. based on networking) as well as diversified constraints on location and access rights. Economics and pricing of resources can also be considered by the Foggy model in a near future. The ability of Foggy to find a trade-off between infrastructure owners' and tenants' needs, in terms of efficient and optimized use of the infrastructure while satisfying the application requirements, is demonstrated through three use cases in the video surveillance and vehicle tracking contexts., The article has been accepted for publication by IEEE