Professor Ram Durairajan, University of Oregon, Joins Link Oregon as Chief Scientist for Sabbatical
2024-25 Engagement to Focus on Exploration of New Network Topologies, Machine Learning for Network Security, and Data-Driven Frameworks to Identify Optimal Opportunities for Middle-Mile Deployments
Link Oregon is excited to have recognized network and systems researcher Ram Durairajan from the University of Oregon join us as Chief Scientist during his 2024-2025 sabbatical year. Ramakrishnan (Ram) Durairajan is an Associate Professor in the School of Computer and Data Sciences, and co-directs the Oregon Networking Research Group (ONRG) at the University of Oregon.
We spoke with Ram recently about what motivated him to join our organization, his views on the intersection of academic research, public policy and the private sector, and the critical urgency for sustainable and climate-resilient infrastructure. Excerpts from our discussion below.
“The need to move away from today’s static configurations and for networks to become more flexible and adaptable has been well established. While technologies like network functions virtualization and software-defined networking have brought changes at the end point, the optical layer hasn’t been touched. Programmatically adapting even a small portion of a topology (e.g. one small link) can offer dramatic benefits for traffic engineering and even DDOS attack mitigation.”
Professor Ram Durairajan School of Computer and Data Sciences University of Oregon
LO: Oregon has a vibrant high-tech private sector offering any number of interesting opportunities to an academic researcher on sabbatical. What made you choose a non-profit like Link Oregon?
RD: You’re right. I could very easily have chosen to work in the private sector during my sabbatical year. But I was looking for meaningful work that could have broader (statewide) impact. Working with Link Oregon gives me a unique vantage point, where I can see the intersections between university research in emerging networking technologies and approaches, public broadband policy and private sector investments while operating as part of a mission-driven Research and Education Network (REN) such as Link Oregon, which acts as the connective tissue across the Oregon broadband ecosystem. That breadth of landscape would be harder for me to access in a private sector role. I was looking for philosophical alignment and I found it in Link Oregon’s mission.
LO: Where do you see the gap/opportunity in terms of public/private/academic partnerships as they relate to broadband deployments in Oregon today?
RD: There is a significant opportunity to bridge the gap between academic research, public policy, and private sector capabilities. Specifically, collaborations can focus on leveraging academic expertise in data analysis — e.g., using machine learning (ML) — to optimize broadband deployments and close the digital divide. Since closing the digital divide is a “multi-stakeholder problem,” public-private-academic partnerships can enhance the deployment of fiber connectivity, ensure effective use of federal resources like BEAD (the federally supported Broadband Equity, Access and Deployment program through the Oregon Broadband Office) funding, and address persistent gaps in broadband access. By aligning goals and sharing knowledge, these partnerships can create more impactful and sustainable broadband solutions.
LO: Specific to the BEAD (Broadband Equity, Access and Deployment) federal funding as it relates to Oregon, what interests you as a researcher?
RD: I’m really interested in using statistical approaches to see how physical fiber connectivity as it shows up on paper will help bridge the digital divide across our state. If data projects that gaps will continue to exist, I would like to work in collaboration with public/private entities to see how best to close those gaps. I have published previously1 on the need for a techno-economic framework that applies geo-based multi-objective optimization to find the areas with the highest concentration of un/underserved users at the lowest cost to service providers. While not specific to Oregon, that research focused on how such a framework could be used to identify attractive targets for broadband deployment from both cost and impact perspectives.
LO: Can you share a bit about what evolutions you see as critical in network topology, giving the emergence of AI and other data- intensive network workloads as well as rising cybersecurity threats?
RD: The need to move away from today’s static configurations and for networks to become more flexible and adaptable has been well established. While technologies like network functions virtualization and software-defined networking have brought changes at the end point, the optical layer hasn’t been touched. Programmatically adapting even a small portion of a topology (e.g. one small link) can offer dramatic benefits for traffic engineering and even DDOS attack mitigation2. I see the move to optical topology programming3 as inevitable and imminent to improve traffic engineering and augment existing defense capabilities against difficult-to-defend-against forms of DDoS attacks, which are on the rise.
Link Oregon strongly supported House Bill 2049 which led to the funding of the Oregon Cybersecurity Center of Excellence created and operated jointly by Oregon’s three largest universities: Portland State University, Oregon State University and the University of Oregon. Governor Kotek is seen here signing the bill into law last summer. (Image courtesy Prof. BIrol Yesilada, Portland State University.)
LO: Speaking of cyber threats and the use of Machine Learning to mitigate those threats, what are some key considerations to be aware of?
RD: First and foremost, trust. You can’t build a good model without data, and you need to build trust with operators to share data so that robust training models can be built. In networking, no two cyber events might be the same, so training a model to detect unusual patterns would require building a model based on “unknowns.” This is something a foundational (vs traditional) machine learning model would be better at. We could start by collecting and analyzing network data to identify patterns of normal and anomalous behavior. Training foundational models on network data can help us understand benign and malicious patterns and evolve the models to the “unknowns” over time using self-supervised learning. This approach allows the model to adapt to evolving threats and improve its detection capabilities.
Beyond cyber threats, I believe machine learning can have a significant impact on the “detection fatigue” levels of network operators by eliminating false positives, minimizing false alerts and significantly shrinking the number of real issues operators must respond to.
LO: What longer term critical issues keep a future-oriented networking researcher like you up at night?
RD: The resiliency and sustainability of the global Internet relies on our ability to harden our infrastructure—under sea and on land. We are increasingly vulnerable here on multiple fronts with climate-change induced risks—from rising sea levels to tsunamis to earthquakes. Specific to undersea Internet infrastructure4, building redundancy and sustainability globally will require dialogue, deep handshakes and partnerships across diverse constituents—from global governments and policy makers to economists, environmentalists and sustainability advocates. It’s not an easy problem to solve given the unique charters and considerations of each group, but it is a clear and present risk to Internet infrastructure that requires urgent and coordinated global-scale action.
