We conduct research in two broad areas related to the Internet:
Network verification and synthesis.
Managing large networks often involves complex policies (e.g., security, QoS, fault tolerance). On the one hand networks must perform acceptably 99.99% of the time or more, yet studies from Google and others show failures are the norm, and traffic patterns change constantly. Design today is largely ad-hoc. By many estimates a large portion of the IT budget of organizations is driven by the need to manage networks, with configuration errors and design faults often accounting for a large fraction of cyber-attacks. We are designing networks that are formally verifiable and which provide provable performance guarantees despite uncertainty. Our research draws on optimization-theoretic techniques and formal methods. We validate our research on real Internet topologies, traffic data and network configurations, and through large-scale emulation testbeds.
ML-driven optimizations of Internet video, and network support for 360 video/AR/VR.
Many exciting new applications are emerging on the Internet such as 4K video, 360 video, and AR/VR which are both bandwidth intensive, and can only tolerate tens of
milliseconds of latency for good user experience. We are developing novel algorithms and building prototye systems for ensuring a high quality of experience
for these applications despite variable Internet environments (e.g., cellular settings). Our algorithms are often informed by large-scale real-world data sets
(e.g, of video streaming sessions) and cutting edge ML techniques.
Our research has benefited by support from NSF, Cisco, Google, Facebook, NetApp, AT&T, and Microsoft , and many of our projects have involved collaboration with these organizations. Many of the challenges we address are motivated by real-world experience, require insights into operations of networks at scale, are great fun, and can change theworld! After completing a Ph.D, alumni of Purdue ISL are now working on cutting edge network infrastructure at places such as Facebook,
Google, AT&T,Bytedance, as well as in universities and Government Agencies. Purdue ISL has often involved undergraduate students, who have continued
to pursue graduate studies in top universities, or gone on to good industry positions.
Background needed to work in Purdue ISL:
Prospective students must ideally have a strong background in Computer Science or Computer Engineering with extensive
programming (ideally systems programming) background. However, EE students with a willingness/ability to build strong systems programming skills are welcome.
Since some of our research draws on cutting edge techniques in optimization, machine learning, and formal methods, students with an aptitude for math/theory with
an interest in applications to real-world networking problems may also be a good fit. Students are not expected to have background in all areas at the start of
their Ph.D, but successful students have an ability and willingness to develop the needed background for their research problem as their research evolves.
The video below, and this slide deck present an overview of recent research in our group.
November 2021: Our paper, Xatu: Richer Neural Network Based Prediction for Video Streaming was accepted at ACM Sigmetrics 2022.
August 2021: Our paper, Chimera: exploiting UAS flight path information to optimize heterogeneous data transmission was accepted at IEEE ICNP 2021.
July 2021: Russ Shirey defends his Ph.D thesis, and will be joining a US Government Agency. Congratulations, Dr. Shirey!
April 2021: Our paper, Hey Lumi! Using Natural Language for Intent-Based Network Management was accepted at Usenix ATC 2021.
August 2020: Our paper, Pitfalls of data-driven networking: A case study of latent causal confounders in video streaming was presented at ACM SIGCOMM 2020 Workshop on Network Meets AI & ML (NetAI 2020).
August 2020: Our paper, PCF: Provably Resilient Flexible Routing was presented at ACM SIGCOMM 2020.