AI at the network edge promises real-time perception and decision-making in safety-critical domains such as aerial robotics, autonomous vehicles, and 5G-enabled infrastructures. Yet, operating under resource constraints, dynamic, and adversarial conditions exposes edge AI systems to fragility, inefficiency, and security risks that threaten their safe operation. GuardAI, a Horizon Europe project, introduces a framework for resilient and trustworthy edge AI that unites three pillars: adversarial robustness, context-enhanced inference, and security-by-design. Initial project results include a diffusion-based adversarial purification framework optimized for real-time operation, lightweight deep unrolling architectures for LiDAR super-resolution with built-in outlier removal, and robust uncertainty quantification modules to improve confidence calibration. It further develops a context-enhanced inference engine that integrates visual, spatial, and operational context across multi-agent systems, and a risk-aware defense recommender that autonomously selects mitigation strategies based on evolving threat landscapes. Through representative Use Cases, covering monitoring with Unmanned Aerial Vehicle, decentralized 5G network analytics, and secure perception in connected autonomous vehicles, GuardAI demonstrates how robust and adaptive AI can be achieved within stringent edge constraints. Together, these technologies lay the groundwork for a new generation of secure, context-aware, and certifiable AI systems that can be trusted to operate autonomously in the physical world.

The efficient transmission of TCP traffic over OBS networks is a challenging problem, due to the high sensitivity of TCP congestion control mechanism to losses. In this paper, a traffic prediction scheme is proposed that exploits TCP traffic dynamics to optimize the performance of TCP transmission over OBS networks. Due to the TCP flow control mechanism, traffic dynamics can be accurately predicted in at least one RTT-long prediction window. In the proposed scheme, the prediction process is tightly coupled with the burst assembly process since the edge node is capable of inspecting incoming TCP traffic, keeping traffic statistics in parallel to the assembly process. These statistics are then used for traffic predictions. In this way burst size can be predicted and thus in advance reserve the appropriate resources. In this paper, we detail the traffic prediction mechanism and we also provide simulation results to assess its performance.

The efficient transmission of TCP traffic over OBS networks is a challenging problem, due to the high sensitivity of TCP congestion control mechanism to losses. In this work, we propose solutions to this problem, through the burst assembly and scheduling mechanism of OBS. We investigate and propose

In Optical Burst Switching (OBS) networks one of the most challenging problems is to provide service differentiation. The lack of buffering in the intermediate nodes, makes QoS provisioning in OBS networks a difficult problem to address. In this work, we propose a scheduling algorithm that supports

In future transparent optical networks, it is important to consider the impact of physical impairments in the routing and wavelengths assignment process, to achieve efficient connection provisioning. In this paper, we use classical multi-objective optimization (MOO) strategies and particularly genetic algorithms to jointly solve the impairment aware RWA (IA-RWA) problem. Fiber impairments are indirectly considered through the insertion of the path length and the number of common hops in the optimization process. It is shown that blocking is greatly improved, while the obtained solutions truly converge towards the Pareto front that constitutes the set of global optimum solutions. We have evaluated our findings, using an Q estimator tool, that calculates the signal quality of each path analytically.

In this paper, a new service oriented networking paradigm is presented, where network nodes (peers) are self- organized into individual service entities. The key idea relies on the overlay approach, where there exists a virtual service plane, fragmented into self-organized and self-managed entities called islands of service transparency. The islands are formed in an upstream, ad-hoc mode from the non-networking resources (i.e VoD, grid server, etc) towards all ingress routers of the network, using link state advertisements and multi-cost path selection algorithms (i.e residual bandwidth, server capacity, storage, etc). Organization and re-organization of nodes around non-network resources is transparent to end-users, and thus any request within a specific service island is transparently routed to the island's resource for execution. A service proxy is commissioned to resolve service addresses and service attributes to QoS metrics. In this paper, we present the main notations and metrics of the proposed architecture as well as node behavior and potential GMPLS extensions for implementation issues.

In this paper, we study the burstification effect on the TCP synchronization and TCP congestion window mechanism. It is shown that short assembly times are optimal for flows with similar characteristics, while large assembly time provide a higher notion of fairness. In addition, this paper analyzes the synchronization of multiple TCP flows when aggregated together over the same optical bursts. It is shown that there is a strong TCP synchronization effect upon burst losses, in the sense that flows’ windows increase/decrease simultaneously, resulting to a significant variation of outgoing traffic. These deficiencies may be dealt with employing a multi-queue burst assembly scheme with different timers and taking into account the TCP dynamics.

In this paper, we present the architectural design of optical burst-buffers that can truly emulate input queuing and accommodate asynchronous burst operation. The architectural design uses wavelength converters and fixed feed-forward delay lines that are combined to form either a multiple-input buffer or a shared buffer. Both schemes are modular, allowing the logarithmic expansion of buffer size with the number of switching elements (wavelength converters).

This work provides a jitter analysis of size-based burst assembly algorithms and also discusses other burst assembly algorithms that use the packet delay as the assembly threshold to provide a bound on jitter.

In this paper we present a signaling protocol for QoS differentiation suitable for optical burst switching networks. The proposed protocol is a two-way reservation scheme that employs delayed and in-advance reservation of resources. In this scheme delayed reservations may be relaxed, introducing a reservation duration parameter that is negotiated during call setup phase. This feature allows bursts to reserve resources beyond their actual size to increase their successful forwarding probability and is used to provide QoS differentiation. The proposed signaling protocol offers a low blocking probability for bursts that can tolerate the round-trip delay required for the reservations. We present the main features of the protocol and describe in detail timing considerations regarding the call setup and the reservation process. We also describe several methods for choosing the protocol parameters so as to optimize performance and present corresponding evaluation results. Furthermore, we compare the performance of the proposed protocol against that of two other typical reservation protocols, a Tell-and-Wait and a Tell-and-Go protocol.

In this paper, we present a new hybrid switching architecture, termed as SLIP-IN, that combines electronic packet/burst with optical circuit switching. SLIP-IN architecture takes advantages of the pre-transmission idle periods of optical lightpaths and slips into them packets or bursts of packets. In optical circuit switching (wavelength-routing) networks, capacity is immediately hard-reserved upon the arrival of a setup message, but is only used after a round-trip time delay. This idle period is significant for optical multi-gigabit networks and can be used to transmit traffic of a lower class of service. In this paper, we present the main features and dependencies of the proposed hybrid switching architecture, and further we perform a detailed evaluation by conducting network wide simulation experiments on the NSFnet backbone topology. For this purpose, we have developed an extensive network simulator, where the basic features of the architecture were modeled. The extensive network study revealed that SLIP-IN architecture can achieve and sustain an adequate data rate with a finite worst case delay.

We present a detailed performance evaluation of a hybrid optical switching architecture called Overspill Routing in Optical Networks (ORION). The ORION architecture combines wavelength and (electronic) packet switching, so as to obtain the advantages of both switching paradigms. We have developed an extensive network simulator where the basic features of the ORION architecture were modeled, including suitable load-varying sources and edge/core node architectures. Various aspects of the ORION architecture were studied including the routing policies used (i.e. once ORION always ORION and lightpath reentry) and the various options available for the buffer architecture. The complete network study shows that ORION can absorb temporary traffic overloads, as intended, provided sufficient buffering is present.

This papers reports the advances in optical core networks research coordinated in the framework of the e- photon/ONe and e-photon/ONe+ networks of excellence.

We present a 40 Gb/s asynchronous self-routing network and node architecture that exploits bit and packet level optical signal processing to perform synchronization, forwarding and switching. Optical packets are self-routed on a hop-by-hop basis through the network by using stacked optical tags, each representing a specific optical node. Each tag contains control signals for configuring the switching matrix and forwarding each packet to the appropriate outgoing link and onto the next hop. Physical layer simulations are performed, modeling each optical sub-system of the node showing acceptable signal quality and bit error rates. Resource reservation-based signaling algorithms are theoretically modeled for the control plane capable of providing high performance in terms of blocking probability and holding time.

In this paper, we present a comprehensive traffic analysis of the Greek School Network (GSN), a wide area network designed to provide Internet access and services to about 15,000 units of primary and secondary education schools and administration offices. In our analysis, we have used measurements from the PATRAS region node obtained through the Cisco NetFlow and FlowScan tools. We have used classical analysis to obtain protocol and application traffic statistics. Our study revealed that TCP traffic is dominant in the network, while nearly 50% of the outgoing and 37% of the incoming traffic is peer-to-peer (P2P) traffic, with a further 25.6% of traffic using not registered ports and suspected to be P2P as well. Finally, we have also observed a remarkable traffic locality phenomenon in the P2P services, where more than the 90% of the traffic was heading or generated by 50 hosts.

In this paper a performance analysis of the packet scheduling switch is presented. The scheduling switch uses a series of feedforward delays interconnected with elementary optical switches. This series of programmable delay blocks constitutes an optical buffer of depth T, whose purpose is to delay/re-arrange incoming packets that request the same outgoing link so as to resolve or reduce packet contention. Performance results have been obtained for random Bernoulli traffic, Pareto traffic, as well as for smooth traffic with an upper bound of inherent burstiness.

We present an innovative protocol processor component that combines wire-speed processing for low-level, and best effort processing for higher-level protocols. The component is a System-on-Chip that integrates variable size packet buffering, specialised cores for header and field processing, generic RISC cores and scheduling blocks. We focus on the main innovation, the reprogrammable pipeline module, and discuss its internal architecture, optimised to perform field processing on byte streams, as well as protocol processing on complex data structures. Furthermore, we present how modern and new tools were used in system dimensioning, design, and verification phases. The chip is able to handle up to 512K flows organised in individual queues. It embeds 5 custom cores optimised for field processing, 3 typical RISC cores for packet processing and 11 generic and application specific hardware blocks. It's been prototyped in UMC 0.18/spl mu/m CMOS technology in a 1096-pin BGA package and operates at 200MHz for 2.5Gbps links.