Hej! I am Mariano, an Italian researcher based in Sweden.
My Life in Research: Coffee First, Ideas Later
My research journey began during my master's thesis at Roma Tre University, where I worked under the supervision of Prof. Giuseppe Di Battista. I was tasked with developing a distributed version of the Kathará network emulator. After months of hard work (and several technical challenges!), I successfully used Kubernetes alongside a custom CNI plugin to emulate Layer 2 networks (combining VXLAN and EVPN BGP). This initial version of "distributed Kathará" evolved into Megalos, my first research paper, which was accepted at NOMS 2020.
This project sparked my passion for research and led me to pursue a PhD with the Computer Networks group under the guidance of Prof. Di Battista. My doctoral work centered on data centers, with an initial focus on underlay routing protocols such as BGP.
This led to the development of Sibyl, the first framework for testing and evaluating routing protocol implementations in Fat-Tree topologies. Notably, Sibyl was instrumental in assessing early implementations of the Routing in Fat-Trees (RIFT) protocol, including both the open-source version (RIFT-Python) and Juniper Networks’ production-grade version. Sibyl also achieved success from an industrial standpoint: we presented it at two IETF meetings (IETF 107 and IETF 114), published a detailed blog post on the APNIC blog, and attracted interest from other researchers across the field.
While developing Sibyl, we required a highly efficient network emulator capable of deploying scenarios in seconds. At that time, Kathará was still in its early stages and faced several technical limitations, and Megalos was not fully integrated within the framework. But, we needed a unified interface to streamline experiments execution. Together with Lorenzo Ariemma, and later Tommaso Caiazzi, we completely rewrote Kathará to eliminate the previous limitations. We presented this updated version, Kathará 2.0, at NOMS 2020, where it won the Best Demo Award! Since then, Kathará has become my most successful project, and I remain one of its main maintainers. We have presented Kathará at various venues, including RIPE, and today, more than 20 universities worldwide use it to teach computer networks at various levels.
After Sibyl, my focus shifted to programmable switches and high-speed networking in collaboration with the NSLab group at KTH. Together, we began exploring an approach to split packets, allowing the NF to process only headers to boost performance. This project also marked my first experience with the P4 language. Eventually, we gained access to a real Tofino switch, and I was instantly captivated: the ability to write your own code and run it at Tbps on actual hardware was fascinating! After several months of development, this work led to Ribosome, published at NSDI 2023. Ribosome implements a packet processing pipeline designed for advanced NF packet processing at Tbps.
After completing my PhD, I was eager to deepen my research in programmable switches and high-speed networking, so I joined the NSLab group as a postdoc on the ERC-funded "ULTRA" project. During this time, I worked on several projects related to programmable networks, NF acceleration, and the emerging role of LLMs in networking.
My initial focus was on designing a high-speed key-value data structure for programmable data planes without control plane interactions. This effort culminated in Switcharoo, which won the Best Paper Award at CoNEXT 2023! Switcharoo introduces a key-value mechanism capable of handling millions of low-latency insertions directly within the data plane. Another key project I contributed to is FAJITA, published at CoNEXT 2024, which implements an optimized packet processing pipeline that achieves >100Mpps on a single CPU and supports stateful NF service chains with different flow definitions. As interest in LLMs grew during my postdoc, we pioneered the exploration of LLM applications for simplifying network configuration and management. This work led to NetConfEval, also published at CoNEXT 2024, the first benchmark suite designed to assess the capabilities of LLMs for network configuration tasks. NetConfEval has garnered significant attention within the networking community, being also shortlisted for the Best Paper Award at CoNEXT 2024. It also received the prestigious IRTF Applied Networking Research Prize in the 2025 edition, standing out among 69 submissions as one of the top six selections.
On November 2024, I joined RISE (Research Institutes of Sweden) as Senior Researcher. I mainly worked on the SEMLA (Securing Enterprises via Machine-Learning-based Automation) project, which leverages ML/AI to automate vulnerability discovery, secure code generation, and resilient infrastructure design. The efforts resulted in several publications focusing on LLMs and cybersecurity and verifiable LLM-based code generation. In parallel, together with (most of!) the Ribosome team, we further explored how to improve Tbps-scale NF deployments, with a strong emphasis on reducing energy consumption. This effort led to Queue-Mem, to be presented at NSDI 2026, which introduces a general-purpose storage abstraction that leverages existing switch buffer queues to support advanced per-flow stateful NFs at line rate, achieving Tbps throughput with a single ASIC switch and one commodity NF server.
Following the conclusion of the SEMLA project in November 2025, I continue to advance my research agenda by contributing in two projects that push innovation in high-speed networking and applications of LLMs. The first Vinnova-funded project, SAFIR (Secure AI for Intelligent Resilience and Confidentiality in the Cloud), is a collaboration between academia (KTH, RISE) and industry (CanaryBit, Ericsson, Nvidia, RedHat, Saab) that aims to build secure, resilient, and trustworthy cloud-edge infrastructures by enabling trusted execution of AI workloads and secure, AI-driven cloud orchestration. The second project, EdgeWise (Fast and Explainable Next-Generation Robots for Mission Critical Applications), a collaboration between KTH and RISE, focuses on developing the next generation of trustworthy humanoid robots, with an emphasis on responsiveness, reliability, and safety through advanced computing and verification techniques that operate effectively under intermittent connectivity. The project is funded by Digital Futures.
Side Projects That Got Out of Hand (in a Good Way)
Throughout these years, I also took on industry-focused side projects involving Kathará.
Alongside Antonio Prado and Tommaso Caiazzi, I developed ROSE-T, the first tool designed to automate MANRS compliance verification. We unveiled ROSE-T at RIPE 87 and provided an in-depth presentation on the "Between 0x2 Nerds" podcast. The project also caught the attention of the Global Cyber Alliance (GCA), the organization behind MANRS, that gave us the opportunity to present the project in the MANRS Community Meeting 2024.
With the Kathará team (myself, Tommaso, and Lorenzo), we are developing a digital twin for the Namex IXP. Initially, this project was a straightforward replica of the Namex network topology, but soon evolved into a robust solution, with automatic updates based on real-time topology changes and various preliminary onboarding checks. The digital twin was officially presented at NAM 2024, and is now publicly available to Namex members. Namex also highlighted the project in an article on the RIPE NCC Labs blog. The project was also introduced to the broader RIPE community at RIPE 90, featuring an in-depth presentation led by Marta Burocchi (Network Engineer at Namex) and Lorenzo Ariemma (Kathará Developer). Riding the wave of the Namex digital twin's success, other Italian IXPs have shown interest in the project. Notably, we are now also partners of VSIX, an IXP based in Padua, to design and deploy the digital twin for their production network.
