HPSR2026

Hi, this is Andrea.

This June I attended HPSR2026 (High Performance Switching and Routing) in Montreal, Canada. It was a really great experience in the foggy and cloudy Montreal, the Gothic style mixed with finance bros skyscrapers made it a peculiar contrast; in the meantime I also visited a friend of mine doing his Master's thesis at the University of Montreal.

I was surprised by how good the conference turned out to be, mainly due to the banquets fact that the conference followed a single-track format. Basically I didn't have to choose between different parallel sessions, and had to attend every presentation. This made it way easier for me to discover new topics, perspectives and get to know very, very interesting people. You might expect a telco conference about switches and routers to have little in common with hot topics like AI. Well, that could be not further from the truth: AI devastated this field as well I'm joking.

Instead, I discovered the beauty of DCNs, (for example always seen optical switches as something too far from my interests, these words didn't have a shape before), more low level topics (what I am mostly thrilled about) such as RoCE, source routing in Linux, and more infrastructure level problems, such as how network architectures are becoming increasingly aware of GenAI models and tasks. As a pratical example think of serving MoE architectures by physically locating different experts on different machines of the same slice, and now the router component actually becomes a packet router. This has been my best experience at a conference so far.

Finally, I am realizing that even though everyone today is pushing people in thinking that you want to build a startup and do a millionaire exit make tons of money to be happy, (In SF this is the actual mentality, @ Stanford it's even more predominant), I might want to do a PhD but not in Italy, please God, and ultimately play with Big-Tech toys, solving real-world, thrilling problems, that affect millions or even billions of people.

because let's be real, big-tech is the only guy with the big toys, and you might want to play with them

p.s. the work I presented is Federated Intrusion Detection with Key-Value Cached Transformers in 5G RANs with Emilio Paolini, Prof. Valcarenghi and Prof. Andriolli.