Doctoral Thesis Oral Defense - Shawn Shuoshuo Chen
August 3, 2026 10:00AM—12:00PM
Location:
4405 & Zoom
-
Gates and Hillman Centers
Speaker:
SHAWN SHUOSHUO CHEN,
Ph.D. Candidate, Department of Computer Science, Carnegie Mellon University
https://shuoshuc.github.io/
As switch chips approach the bandwidth wall, conventional data center networks that interconnect all servers with static, uniform bandwidth become difficult to scale. Reconfigurable data center networks (RDCNs), enabled by optical switching technologies, emerge as a new architecture to allocate time-varying, nonuniform bandwidth between servers based on real-time application demands. However, the existing network protocol stack is designed with long-held assumptions that are incompatible with this new architecture. As a result, we face various inefficiencies in RDCNs across the transport, routing, and application layers.
In this talk, I will present systematic optimizations across the network stack to unlock high performance in this new architecture. I will start with TDTCP, a transport design tailored for paths whose bandwidth and latency change on reconfiguration timescales. I will then introduce PreciseTE, a memory-efficient weighted multipath routing scheme designed for nonuniform bandwidth. Finally, I will discuss RFold, a machine learning job mapping system that reduces network contention and improves resource utilization by co-adapting application-layer communication and network topology.
Thesis Committee:
Srinivasan Seshan (Co-Chair)
Peter Steenkiste (Co-Chair)
Tim Dettmers
Minlan Yu (Harvard University)
In-person and Zoom
Contact
Matt Stewart