Search

In federated learning, distributed nodes train a local machine learning model and exchange it through a central aggregator. In real environments, these training nodes are heterogeneous in computing capacity and bandwidth, thus their specific characteristics influence the performance of the federated learning process. We propose for such situations the design of a federated learning server that is able to adapt dynamically to the heterogeneity of the training nodes. In experiments with real devices deployed in a wireless mesh network, we observed that the designed adaptive federated learning server successfully exploited the idle times of the fast nodes by assigning them larger training workloads, which led to a higher global model performance without increasing the training time., This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 871582 — NGIatlantic.eu and was partially supported by the Spanish Government under contracts PID2019-106774RB-C21, PCI2019-111851-2 (LeadingEdge CHIST-ERA), PCI2019-111850-2 (DiPET CHIST-ERA), and by national funds through FCT, Fundação para a Ciência e a Tecnologia, Portugal, under project UIDB/50021/2020. The work of C.-H. Liu was supported in part by the U.S. National Science Foundation (NSF) under Award CNS-2006453 and in part by Mississippi State University under Grant ORED 253551-060702. The work of L. Wei is supported in part by the U.S. National Science Foundation (#2006612 and #2150486)., Peer Reviewed, Postprint (author's final draft)

Machine learning is moving to the smallest computing devices. Today machine learning is applied even in tiny IoT microcontroller boards. In the IoT, LoRa is a popular communication technology to connect remote devices with gateways. Still, the confluence of machine learning in microcontrollers and networked LoRa connectivity is not yet fully exploited. In this paper we design a new LoRa connectivity for the Arduino Portenta H7, a recent microcontroller board equipped with embedded sensors suitable for diverse machine learning tasks. With the solution that we found the Arduino Portenta H7 is able to become part of a LoRa mesh network. This capacity increases the Portenta's range of applications. For the vision of distributed machine learning at the tiny edge, we can add with the Portenta an important board to become a smart compute node within a LoRa mesh network., This work was partially supported by the Spanish Government under contracts PID2019-106774RB-C21, PCI2019-111851-2 (LeadingEdge CHIST-ERA), PCI2019-111850-2 (DiPET CHIST-ERA)., Peer Reviewed, Postprint (author's final draft)

Federated learning is a distributed learning technique in which a machine learning model is trained collaboratively among several nodes. While the privacy preservation of the training data is one of the important promises of federated learning, there is also an opportunity to use low capacity devices for machine learning model training by taking advantage of the fact that the training effort is divided among many nodes. In this paper, we conduct experiments with a federated learning network deployed on several low capacity devices connected to a wireless mesh network. The measurements show the hardware capacity and link bandwidth of the clients on the federated learning process. The results suggest that for heterogeneous networks the federated learning clients should be extended with more autonomous decision capacities according to the network and local conditions., This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No. 871582 - NGIatlantic.eu and was partially supported by the Spanish Government under contracts PID2019-106774RB-C21, PCI2019-111851-2 (LeadingEdge CHIST-ERA), PCI2019-111850-2 (DiPET CHIST-ERA). The work of C.-H. Liu was supported in part by the U.S. National Science Foundation (NSF) under Award CNS-2006453 and in part by Mississippi State University under Grant ORED 253551-060702. The work of L. Wei is supported in part by the U.S. National Science Foundation (#2006612 and #2150486)., Peer Reviewed, Postprint (author's final draft)

There is a growing research interest in Federated Learning (FL), a promising approach for data privacy preservation and proximity of training to the network edge, where data is generated. Resource consumption for Machine Learning (ML) training and inference is important for edge nodes, but most of the proposed protocols and algorithms for FL are evaluated by simulations. In this demo paper, we present an environment based on distributed mini-PCs to enable experimental study of FL protocols and algorithms. We have installed low-capacity mini-PCs within a wireless city-level mesh network and deployed container-based FL components on these nodes. We show the deployed FL clients and server at different nodes in the city and demonstrate how an FL experiment can be set and run in a real environment., This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 871582 — NGIatlantic.eu and was partially supported by the Spanish Government under contracts PID2019-106774RB-C21, PCI2019-111851-2 (LeadingEdge CHIST-ERA), PCI2019-111850-2 (DiPET CHIST-ERA). The work of C.-H. Liu was supported in part by the U.S. National Science Foundation (NSF) under Award CNS-2006453 and in part by Mississippi State University under Grant ORED 253551-060702. The work of L. Wei is supported in part by the U.S. National Science Foundation (#2150486 and #2006612). I Koutsopoulos acknowledges support from the CHIST-ERA grant CHIST-ERA-18-SDCDN-004 (GSRI grant number T11EPA4-00056)., Peer Reviewed, Postprint (author's final draft)

Community networks are becoming increasingly popular due to the growing demand for network connectivity in both rural and urban areas. Community networks are owned and managed at the edge by volunteers. Their irregular topology, the heterogeneity of resources and their unreliable behavior claim for advanced optimization methods to place services in the network. In particular, an efficient service placement method is key for the performance of these systems. This work presents the Multi-Criteria Optimal Placement method, a novel and fast two-stage multi-objective method to place services in decentralized community network edge micro-clouds. A comprehensive set of computational experiments is carried out using real traces of Guifi.net, which is the largest production community network worldwide. According to the results, the proposed method outperforms both the random placement method used currently in Guifi.net and the Bandwidth-aware Service Placement method, which provides the best known solutions in the literature, by a mean gap in bandwidth gain of about 53% and 10%, respectively, while it also reduces the number of resources used., This work has been partially supported by the Spanish Ministry of Science, Innovation and Universities (PGC2018-097599-B-100 and PID2019-106774RB-C21), and by the Spanish State Research Agency (AEI) under contracts PCI2019-111850-2 (DiPET CHIST-ERA) and PCI2019-111851-2 (LeadingEdge CHIST-ERA)., Peer Reviewed, Postprint (author's final draft)

The increase of the computing capacity of IoT devices and the appearance of lightweight machine learning frameworks have led to the situation that machine learning can nowadays be run in IoT applications at the network edge. There is an opportunity to implement machine learning algorithms with the more and more computationally powerful edge nodes and using the ever increasing amount of local data coming from nearby sensors. For this purpose, federated learning becomes a promising machine learning approach, where a machine learning model is trained by various nodes using their local data. For performing practical federated learning experiments, we have built a testbed deployed within a wireless city mesh network with geographically distributed low capacity devices. We describe the testbed implementation and show its potential to experimentally study federated learning protocols and algorithms in real edge environments., This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 871582 — NGIatlantic.eu and was partially supported by the Spanish Government under contracts PID2019-106774RBC21, PCI2019-111850-2 (DiPET CHIST-ERA), PCI2019-111851-2 (LeadingEdge CHIST-ERA). The work of C.-H. Liu was supported in part by the U.S. National Science Foundation (NSF) under Award CNS-2006453 and in part by Mississippi State University under Grant ORED 253551-060702. The work of L. Wei is supported in part by the U.S. National Science Foundation (#2006612 and #2150486)., Peer Reviewed, Postprint (published version)

Inference with trained machine learning models is now possible with small computing devices while only a few years ago it was run mostly in the cloud only. The recent technique of Federated Learning offers now a way to do also the training of the machine learning models on small devices by distributing the computing effort needed for the training over many distributed machines. But, the training on these low-capacity devices takes a long time and often consumes all the available CPU resource of the device. Therefore, for Federated Learning to be done by low-capacity devices in practical environments, the training process must not only target for the highest accuracy, but also on reducing the training time and the resource consumption. In this paper, we present an approach which uses a dynamic epoch parameter in the model training. We propose the BePOCH (Best Epoch) algorithm to identify what is the best number of epochs per training round in Federated Learning. We show in experiments with medical datasets how with the BePOCH suggested number of epochs, the training time and resource consumption decreases while keeping the level of accuracy. Thus, BePOCH makes machine learning model training on low-capacity devices more feasible and furthermore, decreases the overall resource consumption of the training process, which is an important asnect towards greener machine learning techniques., This work was partially funded by the Spanish Government under contracts PID2019-106774RB-C21, PCI2019-111850- 2 (DiPET CHIST-ERA), PCI2019-111851-2 (LeadingEdge CHIST-ERA), and the Generalitat de Catalunya as Consolidated Research Group 2017-SGR-990. Suport was given also by the Agency for Electronic Communications (AEK) of North Macedonia., Peer Reviewed, Postprint (author's final draft)

Universal connectivity is still unavailable or expensive for half of the global population, despite being critical for social participation. The deployment of crowdsourced networking infrastructures creates an opportunity for local development, where anyone can deploy a new device. In such infrastructures connectivity offer can expand incrementally and be sustainable through investment and fees resulting from the demand and consumption of content and services, including Internet access, that compensate the cost of the underlying network. While routing coordinates network data flows, economic flows can be coordinated by smart contracts built over a local distributed ledger. We define crowdsourced networks, the concept, architecture, and implementation using a local Ethereum PoA blockchain with Solidity smart contracts that compensate the data traffic contribution and consumption recorded by a traffic monitoring system, on a wireless mesh network. The prototype software has been validated in a controlled mesh network environment. Functional tests show its ability to account and route economic flows with small resource consumption, and therefore confirms these networks can develop organically by the addition of consumer and provider participants to reach the typical scale of most wireless mesh access networks and deliver networking services that aim to be socially and economically sustainable., Peer Reviewed, Postprint (author's final draft)

The Guifi.net community network has evolved during the past 15 years into a telecommunications infrastructure that offers Internet access to more than 80.000 people. The monitoring system currently in place for this network is lagging behind the growth of the infrastructure, requiring manual intervention and counting several single points of failure. In this paper we present REDEMON, a resilient decentralized monitoring system, hosted on distributed and interconnected edge devices, for a reliable, eventually-consistent monitoring of the Guifi.net network, leveraging CRDT-based data structures implemented on AntidoteDB. We developed the REDEMON system as a prototype featuring resilience, decentralization and automation, in order to replace the legacy monitoring system. To assess the system, this prototype was deployed on resource-constraint edge nodes in the Guifi.net production network and evaluated under realistic conditions. The decentralized assignment mechanism successfully achieves setting the minimum number of monitoring servers per network device that satisfies the established system requirements. Besides, by concentrating the workload on the minimum required number of servers running at their maximum capacity, the remaining devices can idle away, reducing the consumption footprint of the system. With regard to computing resources, we measure a moderate CPU and RAM usage by the monitoring system on low-capacity devices, while we observe that a considerable network traffic is required for achieving a resilient and consistent data storage layer. This resilient and decentralized architecture could lay the basis for other edge applications in the cloud computing domain that need to coordinate over distributed and consistent shared data., This work was supported by the European H2020 framework programme project LightKone (H2020-732505), by the Spanish State Research Agency (AEI) under contracts PCI2019- 111850-2 and PCI2019-111851-2, and the Catalan government AGAUR SGR 990., Peer Reviewed, Postprint (author's final draft)

By leveraging resources from the Fed4Fire+ City-Lab testbed, we design the PiGeon edge computing platform that experiments solution that enable ICN based edge services in wireless mesh networks (WMNs). PiGeon combines into a platform several trends in edge computing namely the ICN (Information-Centric Networking), the containerization of services exemplified by Docker, novel service placement algorithms and the increasing availability of energy efficient but still powerful hardware at user premises (Raspberry Pi, mini-PCs, and enhanced home gateways). We underpin the PiGeon platform with Docker container-based service that can be seamlessly delivered, cached and deployed at the network edge. The core of the PiGeon platform is the Decision Engine making a decision on where and when to deploy a service instance to satisfy the service requirements while considering the network status and available hardware resources. We collect network data from a real citywide mesh network such as CityLab FIRE testbed located at the city of Antwerp, Belgium. The collected data is used to feed our service placement heuristic within the PiGeon platform. Through a real deployment in CityLab testbed, we show that our service placement heuristic improves the response time up to 37% for stateful services (Web2.0 service). Apart from improving the QoS for end-users, our results show that ICN plays a key role in improving the service delivery time as well as reducing the traffic consumption in WMNs. The overall effect of ICN in our platform is that most content and service delivery requests can be satisfied very close to the client device, many times just one hop away, decoupling QoS from intra-network traffic and origin server load., This work was supported by the European H2020 framework programme project Fed4Fire+ (732638), by the Spanish State Research Agency (AEI) under contracts PCI2019-111850-2 and PCI2019-111851-2, and the Catalan government AGAUR SGR 990., Peer Reviewed, Postprint (author's final draft)

The growing demand for network connectivity has boosted the number of community networks (CNs). CNs are decentralized and self-organized communication networks owned and managed at the edge by volunteers. Due to the heterogeneity of edge node characteristics, high software and hardware diversity, irregular topology and unreliable behavior of the network, the performance of its services varie depending on where they are hosted. These characteristics of CNs and edge platforms running on them require of advanced simulation-optimization methods to place services. In this context, we propose a simheuristic algorithm to address this stochastic problem. The core of this approach relies on a multi-start metaheuristic with a multi-objective optimization method. Our approach combines Monte Carlo simulation and the multi-criteria optimal placement heuristic, The method is tested using real traces of Guifi.net CN, which is considered to be largest CN worldwide., This work has been partially supported by the Spanish Ministry of Science, Innovation and Universities under contract PGC2018-097599-B-100 and by the Spanish State Research Agency (AEI) under contracts PCI2019-111850-2 and PCI2019-111851-2., Peer Reviewed, Postprint (author's final draft)

This is the peer reviewed version of the following article: Kabbinale, AR, Dimogerontakis, E, Selimi, M, et al. Blockchain for economically sustainable wireless mesh networks. Concurrency Computat Pract Exper. 2020; 32:e5349, which has been published in final form at https://doi.org/10.1002/cpe.5349. This article may be used for non-commercial purposes in accordance with Wiley Terms and Conditions for Self-Archiving., Decentralization, in the form of mesh networking and blockchain, two promising technologies, is coming to the telecommunications industry. Mesh networking allows wider low-cost Internet access with infrastructures built from routers contributed by diverse owners, whereas blockchain enables transparency and accountability for investments, revenue, or other forms of economic compensations from sharing of network traffic, content, and services. Crowdsourcing network coverage, combined with crowdfunding costs, can create economically sustainable yet decentralized Internet access. This means that every participant can invest in resources and pay or be paid for usage to recover the costs of network devices and maintenance. While mesh networks and mesh routing protocols enable self-organized networks that expand organically, cryptocurrencies and smart contracts enable the economic coordination among network providers and consumers. We explore and evaluate two existing blockchain software stacks, Hyperledger Fabric (HLF) and Ethereum geth with Proof of Authority (PoA) intended as a local lightweight distributed ledger, deployed in a real city-wide production mesh network and in laboratory network. We quantify the performance and bottlenecks and identify the current limitations and opportunities for improvement to serve locally the needs of wireless mesh networks, without the privacy and economic cost of relying on public blockchains., This paper has been supported by the AmmbrTech Group, the Spanish Government TIN2016‐77836‐C2‐2‐R and the European Community H2020 Programme netCommons (H2020‐688768). The authors would like to thank the people from the Guifi.net (Guifi‐Sants) community network for hosting the servers and supporting the experiments., Peer Reviewed, Postprint (author's final draft)

We present the case of monitoring a decentralized and crowdsourced network infrastructure, that needs to be monitored over geographically distributed devices at the network edge. It is a characteristic of the target environment that both, the infrastructure to be monitored and the hosts where the monitoring system runs, change over time, and network partitions may happen. The proposed monitoring system is decentralized, and monitoring servers coordinate their actions through an eventually consistent data storage layer deployed at the network edge. We developed a proof-of-concept implementation, which leverages CRDT-based data types provided by AntidoteDB. Our evaluation focuses on the understanding of the continuously updated mapping of monitoring server to network devices, specifically on the effects of different policies for each individual monitoring server to decide on which and how many network devices to monitor. One of the policies is experimented by means of a deployment on 8 real nodes, leveraging the data replication of AntidoteDB in a realistic setting. The observed effects of the different policies are interpreted from the point of view of the trade-off between resource consumption and redundancy., This work was supported by the European H2020 framework programme project LightKone (H2020-732505), by the Spanish government contract TIN2016-77836-C2-2-R and PID2019-106774RB-C21 by the Catalan government contract AGAUR SGR 990., Peer Reviewed, Postprint (author's final draft)

Community-built telecommunication networks such as Guifi.net demonstrate how end users can actively collaborate in the self-provision of network services, for instance by operating a self-organized distributed monitoring system. Network monitoring is performed by many small servers at the users' premises but data are only accessible via a centralized interface. Besides, due to network partitions and churn of the monitoring servers, failures in the monitoring system are frequent, leaving parts of the network unmonitored. Distributed databases are a promising solution for data replication under network partition condition, but they suffer from a trade-off between data consistency and availability. Furthermore, these databases are used in data centers with abundant computing resources, not in light edge networks. In this work we present DIMON, a reliable edge-based, eventually-consistent monitoring system that leverages CRDT-based data structures implemented in AntidoteDB. Conflict-free replicated data types (CRDTs) are able to converge to a consistent state in environments with network partitions as those found in edge networks. Our results give insights on the load of AntidoteDB on edge devices under different scenarios of read and write operations. The experiments carried out in a production network with a real system implemented contribute to the research community's knowledge about the available technologies for a consistent replicated data storage layer to support edge computing clouds., This work was supported by the European H2020 framework programme project LightKone (H2020-732505), by the Spanish government under contract TIN2016-77836-C2-2-R and the Catalan government AGAUR SGR 990., Peer Reviewed, Postprint (author's final draft)

© 2019 Elsevier. This manuscript version is made available under the CC-BY-NC-ND 4.0 license http://creativecommons.org/licenses/by-nc-nd/4.0, Edge computing is radically shaping the way Internet services are run by enabling computations to be available close to the users - thus mitigating the latency and performance challenges faced in today’s Internet infrastructure. Emerging markets, rural and remote communities are further away from the cloud and edge computing has indeed become an essential panacea. Many solutions have been recently proposed to facilitate efficient service delivery in edge data centers. However, we argue that those solutions cannot fully support the operations in Community Mesh Networks (CMNs) since the network connection may be less reliable and exhibit variable performance. In this paper, we propose to leverage lightweight virtualisation, Information-Centric Networking (ICN), and service deployment algorithms to overcome these limitations. The proposal is implemented in the PiCasso system, which utilises in-network caching and name based routing of ICN, combined with our HANET (HArdware and NETwork Resources) service deployment heuristic, to optimise the forwarding path of service delivery in a network zone. We analyse the data collected from the Guifi.net Sants network zone, to develop a smart heuristic for the service deployment in that zone. Through a real deployment in Guifi.net, we show that HANET improves the response time up to 53% and 28.7% for stateless and stateful services respectively. PiCasso achieves 43% traffic reduction on service delivery in our real deployment, compared to the traditional host-centric communication. The overall effect of our ICN platform is that most content and service delivery requests can be satisfied very close to the client device, many times just one hop away, decoupling QoS from intra-network traffic and origin server load., Peer Reviewed, Postprint (author's final draft)

The digital world demands a network infrastructure to supply connectivity to any participant anywhere. Sustainable networks require balanced value flows. Value is connectivity delivered at a material and service cost to compensate, involving diverse participants, ranging from consumers to providers, such as last mile access, regional transport, Internet carriers, or content providers. We focus on the case of wireless mesh networks that deliver connectivity through access points and a mesh network that routes traffic to Internet gateways, provisioned by several device owners and service operators [1, 2, 3]. The presented work is motivated by the need for balance and automation among services delivered, costs and incentives for participation in these decentralised networks. This balance is key for achieving extensible network infrastructures that can deliver widespread availability of Internet connectivity with minimal barriers of entry., This paper has been partially supported by the AmmbrTech Group, the Spanish government TIN2016-77836-C2-2-R and the Catalan government AGAUR SGR 990., Peer Reviewed, Postprint (author's final draft)

Cloud computing is moving from data centers to the network edge. Edge computing introduces lightweight computing devices closer to where data are produced. Local processing enables faster response times for cloud-based services. In this approach, however, cloud elasticity still lies in the data center and not at the edge, and an edge device cannot run services locally beyond its capacity. In this paper we present microclouds as an end user-managed infrastructure built from low-cost home servers leveraging PaaS and SaaS at the network edge. We discuss the following aspects: 1) the microcloud platform architecture and the facilities it offers to the users to manage their services and applications, 2) support services to enable the microcloud provision, and 3) of a number of decentralized applications and their potential to support service provision in microclouds., This work was supported by the European H2020 framework programme project LightKone (H2020-732505), by the Spanish government contract TIN2016-77836-C2-2-R, and by the Catalan government contract AGAUR SGR 990., Peer Reviewed, Postprint (author's final draft)

Community networks (CNs) have gained momentum in the last few years with the increasing number of spontaneously deployed WiFi hotspots and home networks. These networks, owned and managed by volunteers, offer various services to their members and to the public. While Internet access is the most popular service, the provision of services of local interest within the network is enabled by the emerging technology of CN micro-clouds. By putting services closer to users, micro-clouds pursue not only a better service performance, but also a low entry barrier for the deployment of mainstream Internet services within the CN. Unfortunately, the provisioning of these services is not so simple. Due to the large and irregular topology, high software and hardware diversity of CNs, a “careful” placement of micro-clouds services over the network is required to optimize service performance. This paper proposes to leverage state information about the network to inform service placement decisions, and to do so through a fast heuristic algorithm, which is critical to quickly react to changing conditions. To evaluate its performance, we compare our heuristic with one based on random placement in Guifi.net, the biggest CN worldwide. Our experimental results show that our heuristic consistently outperforms random placement by 2x in bandwidth gain. We quantify the benefits of our heuristic on a real live video-streaming service, and demonstrate that video chunk losses decrease significantly, attaining a 37% decrease in the packet loss rate. Further, using a popular Web 2.0 service, we demonstrate that the client response times decrease up to an order of magnitude when using our heuristic. Since these improvements translate in the QoE (Quality of Experience) perceived by the user, our results are relevant for contributing to higher QoE, a crucial parameter for using services from volunteer-based systems and adapting CN micro-clouds as an eco-system for service deployment., Peer Reviewed, Postprint (published version)