Federated learning (FL) is a promising approach for edge/IoT-based distributed machine learning, where both privacy and bandwidth efficiency are essential. However, as time progresses, edge/IoT-based FL faces challenges such as unpredictable concept drift, leading to model performance degradation and the need for frequent retraining. To address these challenges, we propose a federated learning framework designed for heterogeneous IoT devices, capable of handling continuous data distribution changes while accounting for limited storage resources. Our framework introduces a server-side drift detection method to minimize bandwidth usage and optimize retraining times, conserving IoT device resources. We also present an efficient storage management strategy to mitigate catastrophic forgetting by selectively managing incoming data streams within device constraints. Additionally, we develop an exemplar-based online continual learning algorithm that leverages class prototypes in the deep feature space to further combat catastrophic forgetting. We evaluate our framework on image classification tasks using ImageNet and CIFAR-100 datasets across four model architectures, demonstrating significant improvements in adaptation to concept drift and long-term performance stability compared to baseline FL approaches.

The current extensive deployment of IoT devices, crucial for enhancing smart computing applications in diverse domains, necessitates the utilization of essential 5G features, notably network slicing, to ensure the provision of distinct and reliable services. However, the voluminous, dynamic, and varied nature of IoT traffic introduces complexities in network flow classification, traffic analysis, and the accurate determination of network requirements. These complexities pose a significant challenge in effectively provisioning 5G network slices across various applications. To address this, we propose an innovative approach for network traffic classification, comprising a pipeline that integrates Principal Component Analysis (PCA) with KMeans clustering and the Hellinger distance measure. The application of PCA as the initial step effectively reduces the dimensionality of the data while retaining most of the original information, which significantly lowers the computational demands for the subsequent KMeans clustering phase. KMeans, an unsupervised learning method, eliminates the labor-intensive and error-prone process of data labeling. Following this, a Hellinger distance-based recursive KMeans algorithm is employed to merge similar clusters, aiding in the determination of the optimal number of clusters. This results in final clustering outcomes that are both compact and intuitively interpretable, overcoming the inherent limitations of the traditional KMeans algorithm, such as its sensitivity to initial conditions and the requirement for manually specifying the number of clusters. An evaluation of our method using a real-world IoT dataset has shown that our pipeline can efficiently represent the dataset in three distinct clusters. The characteristics of these clusters can be readily understood and directly correlated with various types of network slices in the 5G network, demonstrating the efficacy of our approach in managing the complexities of IoT traffic for 5G network slice provisioning.

Industry 4.0 is leading factories to undergo a significant transformation, where automation is achieved through the use of modern smart technologies, such as 5G & beyond $(5 \mathrm{G}+)$ network and digital twins. Yet, many Industrial Internet of Things (IIoT) applications, including smart factories and robotic repair, present challenges in delivering dedicated and real-time network services between the physical world entities and their digital twins due to the different network requirements of each sub tasks of the applications. Although 5G+ networks can provide high-speed, low-latency, and reliable network services, managing and optimizing the network resources in real-time remains complex and time-consuming. To address these challenges, this paper proposes solutions to manage and optimize $5 \mathrm{G}+$ network resources in real-time, and deliver dynamic and real time network requirements of multi-task digital twin applications.

Existing massive deployments of IoT devices in support of smart computing applications across a range of domains must leverage critical features of 5G, such as network slicing, to receive differentiated and reliable services. However, the voluminous, dynamic, and heterogeneous nature of IoT traffic imposes complexities on the problems of network flow classification, network traffic analysis, and accurate quantification of the network requirements, thereby making the provisioning of 5G network slices across the application mix a challenging problem. To address these needs, we propose a novel network traffic classification approach that consists of a pipeline that combines Principal Component Analysis (PCA), with KMeans clustering and Hellinger distance. PCA is applied as the first step to efficiently reduce the dimensionality of features while preserving as much of the original information as possible. This significantly reduces the runtime of KMeans, which is applied as the second step. KMeans, being an unsupervised approach, eliminates the need to label data which can be cumbersome, error-prone, and time-consuming. In the third step, a Hellinger distance-based recursive KMeans algorithm is applied to merge similar clusters toward identifying the optimal number of clusters. This makes the final clustering results compact and intuitively interpretable within the context of the problem, while addressing the limitations of traditional KMeans algorithm, such as sensitivity to initialization and the requirement of manual specification of the number of clusters. Evaluation of our approach on a real-world IoT dataset demonstrates that the pipeline can compactly represent the dataset as three clusters. The service properties of these clusters can be easily inferred and directly mapped to different types of slices in the 5G network.

Virtual topologies in peer-to-peer networks can reduce the traffic consumed by altering the logical connectivity of peers without altering the underlying network. However, such sparsely connected virtual topologies do not focus on the needs for smart grid applications, which is information dissemination throughout the network, and in turn degrade the performance of distributed control algorithms running on peer-to-peer networks. This paper provides a flexible solution for application developers to prototype and deploy different virtual topologies that balances these trade-offs. First, it introduces a configurable virtual communication topology framework, TopLinkMgr, which enables users to specify any chosen connectivity configuration and deploy peer-to-peer applications using it. Second, it proposes a novel fault-tolerant self-adaptive virtual topology management algorithm, Bounded Path Dissemination, that can ensure the dissemination of information to all peers within a specified number of hops. Experiments show that the algorithm improves on convergence speed and accuracy over state-of-the-art methods and is also robust against node failures while consuming significantly less communication bandwidth.

This paper addresses the challenges of delivering fine-grained Quality of Service (QoS) and communication determinism over 5G wireless networks for real-time and autonomous needs of Industrial Internet of Things (IIoT) applications while effectively sharing network resources. Specifically, this work presents DANSM, a software-defined, dynamic and autonomous network slice management middleware for 5G-based IIoT use cases, such as adaptive robotic repair. The novelty of our approach lies in (1) the use of multiple M/M/1 queues to formulate a 5G network resource scheduling optimization problem comprising service-level and system-level objectives; (2) the design of a heuristics-based solution to overcome the NP-hard properties of this optimization problem, and (3) the implementation of a software-defined solution that incorporates the heuristics to dynamically and autonomously provision and manage 5G network slices that deliver predictable communications to IIoT use cases. Empirical studies evaluating DANSM on our testbed comprising a Free5GC-based core and UERANSIM-based simulations reveal that the software-defined DANSM solution can efficiently balance the traffic load in the data plane thereby reducing the end-to-end response time and improve the service performance by completing 34% more subtasks than a Modified Greedy Algorithm (MGA), 64% more subtasks than First Fit Descending (FFD) and 22% more subtasks than Best Fit Descending (BFD) approaches all while minimizing operational costs.

Deep Learning (DL) model-based AI services are increasingly offered in a variety of predictive analytics services such as computer vision, natural language processing, speech recognition. However, the quality of the DL models can degrade over time due to changes in the input data distribution, thereby requiring periodic model updates. Although cloud data-centers can meet the computational requirements of the resource-intensive and time-consuming model update task, transferring data from the edge devices to the cloud incurs a significant cost in terms of network bandwidth and are prone to data privacy issues. With the advent of GPU-enabled edge devices, the DL model update can be performed at the edge in a distributed manner using multiple connected edge devices. However, efficiently utilizing the edge resources for the model update is a hard problem due to the heterogeneity among the edge devices and the resource interference caused by the colocation of the DL model update task with latency-critical tasks running in the background. To overcome these challenges, we present Deep-Edge, a load- and interference-aware, fault-tolerant resource management framework for performing model update at the edge that uses distributed training. This paper makes the following contributions. First, it provides a unified framework for monitoring, profiling, and deploying the DL model update tasks on heterogeneous edge devices. Second, it presents a scheduler that reduces the total re-training time by appropriately selecting the edge devices and distributing data among them such that no latency-critical applications experience deadline violations. Finally, we present empirical results to validate the efficacy of the framework using a real-world DL model update case-study based on the Caltech dataset and an edge AI cluster testbed.

Services hosted in multi-tenant cloud platforms often encounter performance interference due to contention for non-partitionable resources, which in turn causes unpredictable behavior and degradation in application performance. To grapple with these problems and to define effective resource management solutions for their services, providers often must expend significant efforts and incur prohibitive costs in developing performance models of their services under a variety of interference scenarios on different hardware. This is a hard problem due to the wide range of possible co-located services and their workloads, and the growing heterogeneity in the runtime platforms including the use of fog and edge-based resources, not to mention the accidental complexities in performing application profiling under a variety of scenarios. To address these challenges, we present FECBench (Fog/Edge/Cloud Benchmarking), an open source framework comprising a set of 106 applications covering a wide range of application classes to guide providers in building performance interference prediction models for their services without incurring undue costs and efforts. Through the design of FECBench, we make the following contributions. First, we develop a technique to build resource stressors that can stress multiple system resources all at once in a controlled manner, which helps to gain insights into the impact of interference on an application's performance. Second, to overcome the need for exhaustive application profiling, FECBench intelligently uses the design of experiments (DoE) approach to enable users to build surrogate performance models of their services. Third, FECBench maintains an extensible knowledge base of application combinations that create resource stresses across the multi-dimensional resource design space. Empirical results using real-world scenarios to validate the efficacy of FECBench show that the predicted application performance has a median error of only 7.6% across all test cases, with 5.4% in the best case and 13.5% in the worst case.

Fog/'Edge computing is increasingly used to support a wide range of latency-sensitive Internet of Things (IoT) applications due to its elastic computing capabilities that are offered closer to the users. Despite this promise, IoT applications with user mobility face many challenges since offloading the application functionality from the edge to the fog may not always be feasible due to the intermittent connectivity to the fog, and could require application migration among fog nodes due to user mobility. Likewise, executing the applications exclusively on the edge may not be feasible due to resource constraints and battery drain. To address these challenges, this paper describes URMILA, a resource management middleware that makes effective tradeoffs between using fog and edge resources while ensuring that the latency requirements of the IoT applications are met. We evaluate URMILA in the context of a real-world use case on an emulated but realistic IoT testbed.

Elastic auto-scaling in cloud platforms has primarily used horizontal scaling by assigning application instances to distributed resources. Owing to rapid advances in hardware, cloud providers are now seeking vertical elasticity before attempting horizontal scaling to provide elastic auto-scaling for applications. Vertical elasticity solutions must, however, be cognizant of performance interference that stems from multi-tenant collocated applications since interference significantly impacts application quality-of-service (QoS) properties, such as latency. The problem becomes more pronounced for latency-sensitive applications that demand strict QoS properties. Further exacerbating the problem are variations in workloads, which make it hard to determine the right kinds of timely resource adaptations for latency-sensitive applications. To address these challenges and overcome limitations in existing offline approaches, we present an online, data-driven approach which utilizes Gaussian Processes-based machine learning techniques to build runtime predictive models of the performance of the system under different levels of interference. The predictive online models are then used in dynamically adapting to the workload variability by vertically auto-scaling co-located applications such that performance interference is minimized and QoS properties of latency-sensitive applications are met.

Despite the widespread use of server virtualization technologies in cloud data centers, system administrators experience multiple challenges in configuring the hypervisor's scheduler parameters to optimize its performance. Manually tuning the scheduler's parameters is a common practice, however, this approach is not effective particularly when dealing with dynamically changing workload and resource utilizations on the host machines. This problem becomes even harder if cloud resources are overbooked while hosting both latency-sensitive and batched applications. To address these issues, this paper presents iTune, which is a framework for engineering the performance of a hypervisor intelligently via autonomous scheduler configurations. Concretely, iTune optimizes the Xen hypervisor's scheduler configuration parameters autonomously through a three phase process comprising: (1) Discoverer, which monitors and saves the resource usage history of the host machines and groups set of related host machine workloads, (2) Optimizer, where optimum Xen scheduler configuration parameters for each workload cluster are explored by employing a simulated annealing machine learning algorithm, and (3) Observer, where iTune monitors the resource usage of host machines online, classifies them into one of the categories found in the Discoverer phase, and loads the optimum scheduler parameters determined in the Optimizer phase. Experimental results validate our claims.

A large number of modern applications and systems are cloud-hosted, however, limitations in performance assurances from the cloud, and the longer and often unpredictable endto-end network latencies between the end user and the cloud can be detrimental to the response time requirements of the applications, specifically those that have stringent Quality of Service (QoS) requirements. Although edge resources, such as cloudlets, may alleviate some of the latency concerns, there is a general lack of mechanisms that can dynamically manage resources across the cloud-edge spectrum. To address these gaps, this research proposes Dynamic Data Driven Cloud and Edge Systems (D3CES). It uses measurement data collected from adaptively instrumenting the cloud and edge resources to learn and enhance models of the distributed resource pool. In turn, the framework uses the learned models in a feedback loop to make effective resource management decisions to host applications and deliver their QoS properties. D3CES is being evaluated in the context of a variety of cyber physical systems, such as smart city, online games, and augmented reality applications.

Despite the known benefits of hosting cloud-based services, the longer and often unpredictable end-to-end network latencies between the end user and the cloud can be detrimental to the response time requirements of the interactive cloud-hosted applications. Existing efforts that exploit edge/fog technology to migrate services closer to clients in order to improve response times do not fully resolve this problem as they do not focus on performance and interference issues at the migrated locations. This paper proposes INDICES framework that addresses these limitations by providing a novel solution that determines when and to which MDC a service should be migrated to and thus provides the desired performance. Empirical results validating our claims are presented using a setup comprising a centralized cloud and MDCs composed of heterogeneous hardware.

Power and performance tradeoffs are critical and challenging issues faced by cloud service providers (CSPs) while managing their data centers. On the one hand, CSPs strive to reduce power consumption of their data centers to not only decrease their energy costs but to also reduce adverse impact on the environment. On the other hand, CSPs must deliver performance expected by the applications hosted in their cloud in accordance with predefined Service Level Agreements (SLAs). Not doing so will lead to loss of customers and thereby major revenue losses for the CSPs. Addressing these dual set of challenges is hard for the CSPs because power management and performance assurance are conflicting objectives, particularly in the context of multi-tenant cloud systems where multiple virtual machines (VMs) may be hosted on a single physical server. The problem becomes even harder when real-time applications are hosted in these VMs. To address these challenges and make appropriate tradeoffs, we present iPlace, which is an intelligent and tunable power- and performance-aware VM placement middleware. The placement strategy is based on a two-level artificial neural network which predicts (1) CPU usage at the first level, and (2) power consumption and performance of a host machine at the second level that uses the predicted CPU usage. The efficacy of iPlace is evaluated in the context of a VM consolidation algorithm that is applied to running virtual machines and host machines in a private cloud.

There is a growing trend towards migrating applications and services to the cloud. This trend has led to the emergence of different cloud service providers (CSPs), in turn leading to different cost models offered by these CSPs to lease their resources, variabilities in the granularity and specification of resources provided, and heterogeneous APIs offered by the CSPs to the users to program resource requests and deployment for their cloud-hosted services. These challenges make it hard for customers of the cloud to seamlessly transition their services to the cloud or migrate between different CSPs. To address these challenges, this paper presents a solution based on model-driven engineering (MDE). Specifically, we describe the design of the domain-specific modeling languages in our MDE framework and the associated generative mechanisms that address the challenges related to estimating performance and cost to host the services in the cloud, automated deployment and resource management.