Edge computing shifts data processing from centralized servers to the network edge, where data is generated and consumed. This shift is critical to support applications in mobile and Internet of Things (IoT) environments, but it introduces two big challenges: edge nodes have limited computational resources, and network latency between nodes can be significant. Applications in this context are often structured as collections of micro-services, each of which must be strategically placed on edge nodes to minimize network latency, balance workload, and meet Service Level Agreements (SLAs). Additionally, minimizing energy consumption is crucial, which requires limiting the number of active edge nodes. As a result, determining optimal micro-service placement within an edge infrastructure is a complex problem, typically addressed using heuristics capable of producing effective solutions under diverse conditions. To address the inherently conflicting objectives of minimizing end-to-end latency, balancing workload across heterogeneous edge nodes, and minimizing energy consumption in micro-service placement, a multi‐objective optimization effectively exploring the trade-off between power consumption and performances in micro-service edge placement is required. Through a properly modified genetic algorithm that leverages Pareto‐front optimization and selection of active nodes number, in this paper we propose and evaluate a solution that dynamically identifies an appropriate subset of edge nodes and assigns micro-services, consistently delivering high‐quality placement strategies within a limited number of generations, and providing stable performance across a wide range of problem characteristics with minimal parameter tuning.

The Fog computing paradigm makes use of dispersed, diverse, and resource-limited devices located at the network edge to effectively implement Internet of Things (IoT) application services that demand low latency and substantial bandwidth. At the same time, the adoption of microservice-based architectures in the IoT domain is on the rise due to their ability to align with the swift evolution and deployment demands of highly dynamic IoT applications and to elastically scale to fulfill load demands. In complex environments like Fog federations, characterized by highly heterogeneous computing and networking resources, the effective allocation of microservices to available nodes, while ensuring compliance with required Quality of Service (QoS) constraints, represents a significant challenge. In this paper, we present the design and implementation of OptiFog, a comprehensive framework that enables users to model, simulate, and validate microservice placement solutions within a realistic testbed environment. Compared to state-of-the-art approaches, OptiFog offers developers a controlled environment for experimenting with placement solutions while providing the assurance that the resulting deployments will meet the targeted QoS requirements in real-world scenarios, specifically in terms of service execution time and energy consumption of Fog nodes. To demonstrate the feasibility of the proposed approach, we implemented and evaluated a representative use case, involving both sub-optimal and optimal microservice placement, and utilizing real-world microservices drawn from the IoT domain.

The microservices architecture has transformed application development by providing scalability, flexibility, and resilience. However, as organizations scale their infrastructure, deploying microservices across multiple clusters - whether for fault tolerance, geographic distribution, or workload optimization — presents several challenges. Efficient orchestration in these multi-cluster environments is essential to ensure seamless service provisioning, workload distribution, and inter-cluster communication. In this paper, we propose an extension of the OASIS TOSCA standard to support the need of application owners to define deployment schemes that enable them to distribute application components across multiple clusterized environments. To test the viability of the proposed extension, we set up a small-scaled, multi-cluster environment powered with Kubernetes and employed an orchestrator of microservice-based applications that implements the mentioned capability. For the test purpose, a real application from the logistics domain was employed.

Edge computing is a novel paradigm aiming to push computation as close as possible to the data sources and to the end users. This paradigm is extremely important in areas such as the mobile applications and IoT. Key characteristics of edge computing are the limited computational resource of the edge nodes and the presence of non-negligible network delays that can affect the performance. Applications are typically designed as a set of inter-operating micro-services, where each service must be placed on an edge node in order to minimize network latency, while balancing the load distribution over the nodes to avoid the violation the the Service Level Agreements. An additional goal is to minimize energy consumption, meaning that the number of powered-on edge nodes must be kept as low as possible. For these reasons, the problem of micro-service placement over an edge computing infrastructure is complex and must be solved by means of heuristics that must reach suitable solutions under a wide set of operating conditions. We propose a mechanism to solve the placement problem based on genetic algorithms to solve the micro-service placement problem and we analyze the behavior of such heuristic for a wide set of problem characteristics. The proposed algorithm can automatically identify the subset of edge nodes to use and can allocate micro-services to reduce network delays and balance the load. Our experiments demonstrate that the proposed GA is a viable tool to allocate micro-services in edge infrastructures as it can find adequate solutions with a limited number of generations and provides stable performance over a wide set of problem characteristics with limited need for tuning.

Fog computing emerged as a novel approach to deliver micro-services that support innovative applications.This paradigm is consistent with the modern approach to application development, that leverages the composition of small micro-services that can be combined to create value-added applications.These applications typically require the access from distributed data sources, such as sensors located in multiple geographic locations or mobile users.In such scenarios, the traditional cloud approach is not suitable because latency constraints may not be compatible with having time-critical computations occurring on a far away data-center; furthermore, the amount of data to exchange may cause high costs imposed by the cloud pricing model.A layer of fog nodes close to application consumers can host pre-processing and data aggregation tasks that can reduce the response time of latency-sensitive elaboration as well as the traffic to the cloud data-centers.However, the problem of smartly placing micro-services over fog nodes that can fulfill Service Level Agreements is far more complex than in the more controlled scenario of cloud computing, due to the heterogeneity of fog infrastructures in terms of performance of both the computing nodes and inter-node connectivity.In this paper, we tackle such problem proposing a mathematical model for the performance of complex applications deployed on a fog infrastructure.We adapt the proposed model to be used in a genetic algorithm to achieve optimized deployment decisions about the placement of micro-services chains.Our experiments prove the viability of our proposal with respect to meeting the SLA requirements in a wide set of operating conditions.

The Fog Computing paradigm is increasingly seen as the most promising solution to support Internet of Things applications and satisfy their requirements in terms of response time and Service Level Agreements. For these applications, fog computing offers the great advantage of reducing the response time thanks to the layer of intermediate nodes able to perform pre-processing, filtering and other computational tasks. However, the design of a fog computing infrastructure opens new issues concerning the allocation of data flows coming from sensors over the fog nodes, and the choice of the number of the fog nodes to be activated. Many studies rely on a simplified assumption based on a M/M/1 theoretical queuing model to determine the optimal solution for the fog infrastructure design, but such simplification may result in a mismatch between predicted and achieved performance of the model. In this paper, we measure the aforementioned discordance in terms of response time and SLA compliance. Furthermore, we explore the impact of non-Poissonian service models and validate our results by means of simulation. Our experiments demonstrate that the use of M/M/1 model could lead to SLA violations. On the other hand, the use of sophisticated models for the estimation of the response time can avoid this problem.

Cloud Computing is emerging as a major trend in ICT industry. However, as with any new technology, new major challenges lie ahead, one of them concerning the resource provisioning. Indeed, modern Cloud applications deal with a dynamic context that requires a continuous adaptation process in order to meet satisfactory Quality of Service (QoS) but even the most titled Cloud platform provide just simple rule-based tools; the rudimentary autoscaling mechanisms that can be carried out may be unsuitable in many situations as they do not prevent SLA violations, but only react to them. In addition, these approaches are inherently static and cannot catch the dynamic behavior of the application and are unsuitable to manage multi-Cloud/data center deployments required for mission critical services. This situation calls for advanced solutions designed to provide Cloud resources in a predictive and dynamic way. This work presents capacity allocation algorithms, whose goal is to minimize the total execution cost while satisfying some constraints on the average response time of multi-Cloud based applications. This paper proposes a joint load balancing and receding horizon capacity allocation techniques, which can be employed to handle multiple classes of requests. An extensive evaluation of the proposed solution against an Oracle with perfect knowledge of the future and well-known heuristics proposed in the literature is provided. The analysis shows that our solution outperforms the heuristics producing results very close to the optimal ones, and reducing the number of QoS violations (in the worst case QoS constraints violation rate is 4.26 percent versus up to 17.25 percent of other approaches and can easily reduced by roughly a factor of four by exploiting the receding horizon approach). Furthermore, a sensitivity analysis over two different time scales indicates that finer grained time scales are more appropriate for spiky workloads. Analytical results are validated through simulation, which also analyzes the impact of Cloud environment random perturbations. Finally, experiments on a prototype environment demonstrate the effectiveness of the proposed approach under real workloads.

Several fields of application:-Urban applications -

Fog computing infrastructures must support increasingly complex applications where a large number of sensors send data to intermediate fog nodes for processing. As the load in such applications (as in the case of a smart cities scenario) is subject to significant fluctuations both over time and space, load balancing is a fundamental task. In this paper we study a fully distributed algorithm for load balancing based on random probing of the neighbors' status. A qualifying point of our study is considering the impact of delay during the probe phase and analyzing the impact of stale load information. We propose a theoretical model for the loss of correlation between actual load on a node and stale information arriving to the neighbors. Furthermore, we analyze through simulation the performance of the proposed algorithm considering a wide set of parameters and comparing it with an approach from the literature based on random walks. Our analysis points out under which conditions the proposed algorithm can outperform the alternatives.

The success of IoT applications increases the number of online devices and motivates the adoption of a fog computing paradigm to support large and widely distributed infrastructures. However, the heterogeneity of nodes and their connections requires the introduction of load balancing strategies to guarantee efficient operations. This aspect is particularly critical when some nodes are characterized by high communication delays. Some proposals such as the Sequential Forwarding algorithm have been presented in literature to provide load balancing in fog computing systems. However, such algorithms have not been studied for a wide range of working parameters in an heterogeneous infrastructure; furthermore, these algorithms are not designed to take advantage from highly heterogeneous network delays that are common in fog infrastructures. The contribution of this study is twofold: first, we evaluate the performance of the sequential forwarding algorithm for several load and delay conditions; second, we propose and test a delay-aware version of the algorithm that takes into account the presence of highly variable node connectivity in the infrastructure. The results of our experiments, carried out using a realistic network topology, demonstrate that a delay-blind approach to sequential forwarding may determine poor performance in the load balancing when network delay represents a major contribution to the response time. Furthermore, we show that the delay-aware variant of the algorithm may provide a benefit in this case, with a reduction in the response time up to 6%.

A clear trend in the evolution of network-based services is the ever-increasing amount of multimedia data involved. This trend towards big-data multimedia processing finds its natural placement together with the adoption of the cloud computing paradigm, that seems the best solution to cope with the demands of a highly fluctuating workload that characterizes this type of services. However, as cloud data centers become more and more powerful, energy consumption becomes a major challenge both for environmental concerns and for economic reasons. An effective approach to improve energy efficiency in cloud data centers is to rely on traffic engineering techniques to dynamically adapt the number of active servers to the current workload. Towards this aim, we propose a joint computing-plus-communication optimization framework exploiting virtualization technologies, called MMGreen. Our proposal specifically addresses the typical scenario of multimedia data processing with computationally intensive tasks and exchange of a big volume of data. The proposed framework not only ensures users the Quality of Service (through Service Level Agreements), but also achieves maximum energy saving and attains green cloud computing goals in a fully distributed fashion by utilizing the DVFS-based CPU frequencies. To evaluate the actual effectiveness of the proposed framework, we conduct experiments with MMGreen under real-world and synthetic workload traces. The results of the experiments show that MMGreen may significantly reduce the energy cost for computing, communication and reconfiguration with respect to the previous resource provisioning strategies, respecting the SLA constraints.

The growing popularity of the Fog Computing paradigm is driven by the increasing availability of large amount of sensors and smart devices on a geographically distributed area.The scenario of a smart city is a clear example of this trend.As we face an increasing presence of sensors producing a huge volume of data, the classical cloud paradigm, with few powerful data centers that are far away from the data sources, becomes inadequate.There is the need to deploy a highly distributed layer of data processors that filter, aggregate and pre-process the incoming data according to a fog computing paradigm.However, a fog computing architecture must distribute the incoming workload over the fog nodes to minimize communication latency while avoiding overload.In the present paper we tackle this problem in a twofold way.First, we propose a formal model for the problem of mapping the data sources over the fog nodes.The proposed optimization problem considers both the communication latency and the processing time on the fog nodes (that depends on the node load).Furthermore, we propose a heuristic, based on genetic algorithms to solve the problem in a scalable way.We evaluate our proposal on a geographic testbed that represents a smart-city scenario.Our experiments demonstrate that the proposed heuristic can be used for the optimization in the considered scenario.Furthermore, we perform a sensitivity analysis on the main heuristic parameters.

Cloud computing data centers are growing in size and complexity to the point where monitoring and management of the infrastructure become a challenge due to scalability issues. A possible approach to cope with the size of such data centers is to identify VMs exhibiting a similar behavior. Existing literature demonstrated that clustering together VMs that show a similar behavior may improve the scalability of both monitoring andmanagement of a data center. However, available techniques suffer from a trade-off between accuracy and time to achieve this result. Throughout this paper we propose a different approach where, instead of an unsupervised clustering, we rely on classifiers based on deep learning techniques to assigna newly deployed VMs to a cluster of already-known VMs. The two proposed classifiers, namely DeepConv and DeepFFT use a convolution neural network and (in the latter model) exploits Fast Fourier Transformation to classify the VMs. Our proposal is validated using a set of traces describing the behavior of VMs from a realcloud data center. The experiments compare our proposal with state-of-the-art solutions available in literature, demonstrating that our proposal achieve better performance. Furthermore, we show that our solution issignificantly faster than the alternatives as it can produce a perfect classification even with just a few samples of data, making our proposal viable also toclassify on-demand VMs that are characterized by a short life span.

As cloud computing data centers grow in size and complexity to accommodate an increasing number of virtual machines, the scalability of monitoring and management processes becomes a major challenge. Recent research studies show that automatically clustering virtual machines that are similar in terms of resource usage may address the scalability issues of IaaS clouds. Existing solutions provide high clustering accuracy at the cost of very long observation periods, that are not compatible with dynamic cloud scenarios where VMs may frequently join and leave. We propose a novel technique, namely Adaptive Gray Area-based TEchnique (AGATE), that provides accurate clustering results for a subset of VMs after a very short time. This result is achieved by introducing elements of fuzzy logic into the clustering process to identify the VMs with undecided clustering assignment (the so-called gray area), that should be monitored for longer periods. To evaluate the performance of the proposed solution, we apply the technique to multiple case studies with real and synthetic workloads. We demonstrate that our solution can correctly identify the behavior of a high percentage of VMs after few hours of observations, and significantly reduce the data required for monitoring with respect to state-of-the-art solutions.

We target the problem of managing the power states of the servers in a Cloud Data Center (CDC) to jointly minimize the electricity consumption and the maintenance costs derived from the variation of power (and consequently of temperature) on the servers' CPU. More in detail, we consider a set of virtual machines (VMs) and their requirements in terms of CPU and memory across a set of Time Slot (TSs). We then model the consumed electricity by taking into account the VMs processing costs on the servers, the costs for transferring data between the VMs, and the costs for migrating the VMs across the servers. In addition, we employ a material-based fatigue model to compute the maintenance costs needed to repair the CPU, as a consequence of the variation over time of the server power states. After detailing the problem formulation, we design an original algorithm, called Maintenance and Electricity Costs Data Center (MECDC), to solve it. Our results, obtained over several scenarios from a real CDC, show that MECDC largely outperforms two reference algorithms, which instead either target the load balancing or the energy consumption of the servers.

In this paper, we propose an adaptive online energy-aware scheduling algorithm by exploiting the reconfiguration capability of a Virtualized Networked Data Centers (VNetDCs) processing large amount of data in parallel. To achieve energy efficiency in such intensive computing scenarios, a joint balanced provisioning and scaling of the networking-plus-computing resources is required. We propose a scheduler that manages both the incoming workload and the VNetDC infrastructure to minimize the communication-plus-computing energy dissipated by processing incoming traffic under hard real-time constraints on the per-job computing-plus-communication delays. Specifically, our scheduler can distribute the workload among multiple virtual machines (VMs) and can tune the processor frequencies and the network bandwidth. The energy model used in our scheduler is rather sophisticated and takes into account also the internal/external frequency switching energy costs. Our experiments demonstrate that the...

In this paper, we propose a dynamic resource provisioning scheduler to maximize the application throughput and minimize the computing-plus-communication energy consumption in virtualized networked data centers. The goal is to maximize the energy-efficiency, while meeting hard QoS requirements on processing delay. The resulting optimal resource scheduler is adaptive, and jointly performs: i) admission control of the input traffic offered by the cloud provider; ii) adaptive balanced control and dispatching of the admitted traffic; iii) dynamic reconfiguration and consolidation of the Dynamic Voltage and Frequency Scaling (DVFS)-enabled virtual machines instantiated onto the virtualized data center. The proposed scheduler can manage changes of the workload without requiring server estimation and prediction of its future trend. Furthermore, it takes into account the most advanced mechanisms for power reduction in servers, such as DVFS and reduced power states. Performance of the proposed scheduler is numerically tested and compared against the corresponding ones of some state-of-the-art schedulers, under both synthetically generated and measured real-world workload traces. The results confirm the delay-vs.-energy good performance of the proposed scheduler.

Supporting the emerging digital society is creating new challenges for cloud computing infrastructures, exacerbating scalability issues regarding the processes of resource monitoring and management in large cloud data centers. Recent research studies show that automatically clustering similar virtual machines running the same software component may improve the scalability of the monitoring process in IaaS cloud systems. However, to avoid misclassifications, the clustering process must take into account long time series (up to weeks) of resource measurements, thus resulting in a mechanism that is slow and not suitable for a cloud computing model where virtual machines may be frequently added or removed in the data center. In this paper, we propose a novel methodology that dynamically adapts the length of the time series necessary to correctly cluster each VM depending on its behavior. This approach supports a clustering process that does not have to wait a long time before making decisions about the VM behavior. The proposed methodology exploits elements of fuzzy logic for the dynamic determination of time series length. To evaluate the viability of our solution, we apply the methodology to a case study considering different algorithms for VMs clustering. Our results confirm that after just 1 day of monitoring we can cluster without misclassifications up to 80% of the VMs, while for the remaining 20% of the VMs longer observations are needed.

Identification of VMs exhibiting similar behavior can improve scalability in monitoring and management of cloud data centers. Existing solutions for automatic VM clustering may be either very accurate, at the price of a high computational cost, or able to provide fast results with limited accuracy. Furthermore, the performance of most solutions may change significantly depending on the specific values of technique parameters. In this paper, we propose a novel approach to model VM behavior using Mixture of Gaussians (MoGs) to approximate the probability density function of resources utilization. Moreover, we exploit the Kullback-Leibler divergence to measure the similarity between MoGs. The proposed technique is compared against the state of the art through a set of experiments with data coming from a private cloud data center. Our experiments show that the proposed technique can provide high accuracy with limited computational requirements. Furthermore, we show that the performance of our proposal, unlike the existing alternatives, does not depend on any parameter.

Providers of Web-based services can take advantage of many convenient features of cloud computing infrastructures, but they still have to implement request management algorithms that are able to face sudden peaks of requests. We consider distributed algorithms implemented by front-end servers to dispatch and redirect requests among application servers. Current solutions based on load-blind algorithms, or considering just server load and thresholds are inadequate to cope with the demand patterns reaching modern Internet application servers. In this paper, we propose and evaluate a request management algorithm, namely Performance Gain Prediction, that combines several pieces of information (server load, computational cost of a request, user session migration and redirection delay) to predict whether the redirection of a request to another server may result in a shorter response time. To the best of our knowledge, no other study combines information about infrastructure status, user request characteristics and redirection overhead for dynamic request management in cloud computing. Our results show that the proposed algorithm is able to reduce the response time with respect to existing request management algorithms operating on the basis of thresholds.

Social networks have changed the characteristics of the traditional Web and these changes are still ongoing. Nowadays, it is impossible to design valid strategies for content management, information dissemination and marketing in the context of a social network system without considering the popularity of its content and the characteristics of the relations among its users. By analyzing two popular social networks and comparing current results with studies dating back to 2007, we confirm some previous results and we identify novel trends that can be utilized as a basis for designing appropriate content and system management strategies. Our analyses confirm the growth of the two social networks in terms of quantity of contents and numbers of social links among the users. The social navigation is having an increasing influence on the content popularity because the social links are representing a primary method through which the users search and find contents. An interesting novel trend emerging from our study is that subsets of users have major impact on the content popularity with respect to previous analyses, with evident consequences on the possibility of implementing content dissemination strategies, such as viral marketing1.

One of the most critical operations performed in a P2P system is the lookup of a resource. The main issues to be addressed by lookup algorithms are: (1) support for flexible search criteria (e.g., wildcard or multi-keyword searches), (2) effectiveness - i.e., ability to identify all the resources that match the search criteria, (3) efficiency - i.e. low overhead, (4) robustness with respect to node failures and churning. Flood-based P2P networks provide flexible lookup facilities and robust performance at the expense of high overhead, while other systems (e.g. DHT) provide a very efficient lookup mechanism, but lacks flexibility. In this paper, we propose a novel resource lookup algorithm, namely fuzzy-DHT, that solves this trade-off by introducing a flexible and robust lookup criteria based on multiple keywords on top of a distributed hash table algorithm. We demonstrate that the fuzzy-DHT algorithm satisfies all the requirements of P2P lookup systems combining the flexibility of flood-based mechanisms while preserving high efficiency, effectiveness ad robustness.

The last generation of Web is characterized by social networking services where users exchange a growing amount of multimedia content. The impact of these novel services on the underlying Web infrastructures is significantly different from traditional Web-based services and has not yet been widely studied.This paper presents a scalability and bottleneck analysis of a Web system supporting social networking services for different scenarios of user interaction patterns, amount of multimedia content and network characteristics.Our study demonstrates that for some social networking services the user interaction patterns may play a fundamental role in the definition of the bottleneck resource and must be considered in the design of systems supporting novel services.

The great diffusion of Mobile Web-enabled devices allows the implementation of novel personalization, location and adaptation services that will place unprecedented strains on the server infrastructure of the content provider. This paper has a twofold contribution. First, we analyze the five-years trend of Mobile Web-based applications in terms of workload characteristics of the most popular services and their impact on the server infrastructures. As the technological improvements at the server level in the same period of time are insufficient to face the computational requirements of the future Mobile Web-based services, we propose and evaluate adequate resource management strategies. We demonstrate that pre-adaptating a small fraction of the most popular resources can reduce the response time up to one third thus facing the increased computational impact of the future Mobile Web-based services.

The advent of the mobile Web and the increasing demand for personalized contents arise the need for computationally expensive services, such as dynamic generation and on-the- fly adaptation of contents. Providing these services exacerbates the performance issues that have to be addressed by the underlying Web architecture. When performance issues are addressed through geographically distributed Web systems with a large number of nodes located on the network edge, the dispatching mechanism that distributes requests among the system nodes becomes a critical element. In this paper, we investigate how the granularity of re- quest dispatching may affect the performance of a distributed Web system for personalized contents. Through a real prototype, we compare dispatching mechanisms operating at various levels of granularity for different workload and network scenarios. We demonstrate that the choice of the best granularity for request dispatching strongly depends on the characteristics of the workload in terms of heterogeneity and computational requirements. A coarse- grain dispatching is preferable only when the requests have similar computational requirements. In all other instances of skewed workloads, that we can consider more realistic, a fine-grain dispatching augments the control on the node load and allows the system to achieve better performance.

The popularity of ubiquitous Web access requires runtime adaptations of the Web contents. A significant trend in these content adaptation services is the growing amount of personalization required by users. Personalized services are and will be a key feature for the success of the ubiquitous Web, but they open two critical issues: performance and profile management. Issues related to the performance of adaptation services are typically addressed by highly distributed architectures with a large number of nodes located closer to user. On the other hand, the management of user profile must take into account the nature of these data that may contain sensitive information, such as geographic position, navigation history and personal preferences that should be kept private. In this paper, we investigate the impact that a correct profile management has on distributed infrastructures that provide content adaptation services for ubiquitous Web access. In particular, we propose and compare two scalable solutions of adaptation services deployed on the nodes of a two-level topology. We study, through real prototypes, the performance and the constraints that characterize the proposed architectures.

Architectures consisting of multiple cache servers are a popular solution to deal with performance and network resource utilization issues related to the growth of the Web request. Cache cooperation is often carried out through purely hierarchical and flat schemes that suffer from scalability problems when the number of servers increases. We propose, implement and compare the performance of three novel distributed cooperation models based on a two-tier organization of the cache servers. The experimental results show that the proposed architectures are effective in supporting cooperative document lookup and download They guarantee cache hit rates comparable to those of the most performing protocols with a significant reduction of the cooperation overhead Moreover in case of congested network, they reduce the 90-percentile of the system response time up to nearly 30% with respect to the best pure cooperation mechanisms.