The rapidly expanding number of IoT devices is generating huge quantities of data, but public concern over data privacy means users are apprehensive to send data to a central server for Machine Learning (ML) purposes. The easilychanged behaviours of edge infrastructure that Software Defined Networking provides makes it possible to collate IoT data at edge servers and gateways, where Federated Learning (FL) can be performed: building a central model without uploading data to the server. FedAvg is a FL algorithm which has been the subject of much study, however it suffers from a large number of rounds to convergence with non-Independent, Identically Distributed (non-IID) client datasets and high communication costs per round. We propose adapting FedAvg to use a distributed form of Adam optimisation, greatly reducing the number of rounds to convergence, along with novel compression techniques, to produce Communication-Efficient FedAvg (CE-FedAvg). We perform extensive experiments with the MNIST/CIFAR-10 datasets, IID/non-IID client data, varying numbers of clients, client participation rates, and compression rates. These show CE-FedAvg can converge to a target accuracy in up to 6× less rounds than similarly compressed FedAvg, while uploading up to 3× less data, and is more robust to aggressive compression. Experiments on an edge-computing-like testbed using Raspberry Pi clients also show CE-FedAvg is able to reach a target accuracy in up to 1.7× less real time than FedAvg.

Multi-access Edge Computing (MEC) is an emerging paradigm which utilizes computing resources at the network edge to deploy heterogeneous applications and services. In the MEC system, mobile users and enterprises can offload computation-intensive tasks to nearby computing resources to reduce latency and save energy. When users make offloading decisions, the task dependency needs to be considered. Due to the NP-hardness of the offloading problem, the existing solutions are mainly heuristic, and therefore have difficulties in adapting to the increasingly complex and dynamic applications. To address the challenges of task dependency and adapting to dynamic scenarios, we propose a new Deep Reinforcement Learning (DRL) based offloading framework, which can efficiently learn the offloading policy uniquely represented by a specially designed Sequence-to-Sequence (S2S) neural network. The proposed DRL solution can automatically discover the common patterns behind various applications so as to infer an optimal offloading policy in different scenarios. Simulation experiments were conducted to evaluate the performance of the proposed DRL-based method with different data transmission rates and task numbers. The results show that our method outperforms two heuristic baselines and achieves nearly optimal performance.

No abstract

Content caching is a promising approach in edge computing to cope with the explosive growth of mobile data on 5G networks, where contents are typically placed on local caches for fast and repetitive data access. Due to the capacity limit of caches, it is essential to predict the popularity of files and cache those popular ones. However, the fluctuated popularity of files makes the prediction a highly challenging task. To tackle this challenge, many recent works propose learning based approaches which gather the users' data centrally for training, but they bring a significant issue: users may not trust the central server and thus hesitate to upload their private data. In order to address this issue, we propose a Federated learning based Proactive Content Caching (FPCC) scheme, which does not require to gather users' data centrally for training. The FPCC is based on a hierarchical architecture in which the server aggregates the users' updates using federated averaging, and each user performs training on its local data using hybrid filtering on stacked autoencoders. The experimental results demonstrate that, without gathering user's private data, our scheme still outperforms other learning-based caching algorithms such as m--greedy and Thompson sampling in terms of cache efficiency.

Localization techniques are becoming key to add location context to Internet of Things (IoT) data without human perception and intervention. Meanwhile, the newly-emerged Low-Power Wide-Area Network (LPWAN) and 5G technologies have become strong candidates for mass-market localization applications. However, various error sources have limited localization performance by using such IoT signals. This paper reviews the IoT localization system through the following sequence: IoT localization system review -localization data sourceslocalization algorithms -localization error sources and mitigation -localization performance evaluation. Compared to the related surveys, this paper has a more comprehensive and state-of-theart review on IoT localization methods, an original review on IoT localization error sources and mitigation, an original review on IoT localization performance evaluation, and a more comprehensive review of IoT localization applications, opportunities, and challenges. Thus, this survey provides comprehensive guidance for peers who are interested in enabling localization ability in the existing IoT systems, using IoT systems for localization, or integrating IoT signals with the existing localization sensors.

Software-Defined Networking (SDN) is an emerging architecture for the next-generation Internet, providing unprecedented network programmability to handle the explosive growth of big data driven by the popularisation of smart mobile devices and the pervasiveness of content-rich multimedia applications. In order to quantitatively investigate the performance characteristics of SDN networks, several research efforts from both simulation experiments and analytical modelling have been reported in the current literature. Among those studies, analytical modelling has demonstrated its superiority in terms of cost-effectiveness in the evaluation of large-scale networks. However, for analytical tractability and simplification, existing analytical models are derived based on the unrealistic assumptions that the network traffic follows the Poisson process, which is suitable to model nonbursty text data, and the data plane of SDN is modelled by one simplified Single-Server Single-Queue (SSSQ) system. Recent measurement studies have shown that, due to the features of heavy volume and high velocity, the multimedia big data generated by real-world multimedia applications reveals the bursty and correlated nature in the network transmission. With the aim of capturing such features of realistic traffic patterns and obtaining a comprehensive and deeper understanding of the performance behaviour of SDN networks, this article presents a new analytical model to investigate the performance of SDN in the presence of the bursty and correlated arrivals modelled by the Markov Modulated Poisson Process (MMPP). The Quality-of-Service performance metrics in terms of the average latency and average network throughput of the SDN networks are derived based on the developed analytical model. To consider a realistic multiqueue system of forwarding elements, a Priority-Queue (PQ) system is adopted to model the SDN data plane. To address the challenging problem of obtaining the key performance metrics, for example, queue-length distribution of a PQ system with a given service capacity, a versatile methodology extending the Empty Buffer Approximation (EBA) method is proposed to facilitate the decomposition of such a PQ system to two SSSQ systems. The validity of the proposed model is demonstrated through extensive simulation experiments. To illustrate its application, the developed model is then utilised to study the strategy of the network configuration and resource allocation in SDN networks.

No abstract

Resource provisioning for cloud computing necessitates the adaptive and accurate prediction of cloud workloads. However, the existing methods cannot effectively predict the high-dimensional and highly-variable cloud workloads. This results in resource wasting and inability to satisfy service level agreements (SLAs). Since recurrent neural network (RNN) is naturally suitable for sequential data analysis, it has been recently used to tackle the problem of workload prediction. However, RNN often performs poorly on learning longterm memory dependencies, and thus cannot make the accurate prediction of workloads. To address these important challenges, we propose a deep Learning based Prediction Algorithm for cloud Workloads (L-PAW). First, a top-sparse auto-encoder (TSA) is designed to effectively extract the essential representations of workloads from the original high-dimensional workload data. Next, we integrate TSA and gated recurrent unit (GRU) block into RNN to achieve the adaptive and accurate prediction for highly-variable workloads. Using realworld workload traces from Google and Alibaba cloud data centers and the DUX-based cluster, extensive experiments are conducted to demonstrate the effectiveness and adaptability of the L-PAW for different types of workloads with various prediction lengths. Moreover, the performance results show that the L-PAW achieves superior prediction accuracy compared to the classic RNN-based and other workload prediction methods for high-dimensional and highly-variable real-world cloud workloads.