Parallel Data Laboratory Talks - Yiran Lei, Dan Volz, and Chunyan Du
June 11, 2026 12:00PM—1:00PM
Location:
Virtual Presentations
-
Remote Access - Zoom
Speaker:
YIRAN LEI, DAN VOLZ, CHUNYAN DU
► TALK 1
YIRAN LEI
Ph.D. Student, Computer Science Department, Carnegie Mellon University
— The Energy Cost of Execution-idle in GPU Clusters
Modern GPU systems often assume that low visible activity means low power. In this talk, I will introduce execution-idle, a hidden GPU power state where a program remains loaded and visible utilization is near zero, but the GPU still draws elevated power. I will discuss how this behavior appears in real GPU clusters, why it matters for energy efficiency, and what it suggests for future GPU observability and power management.
Yiran Lei is a third-year Ph.D. student in the Computer Science Department at Carnegie Mellon University, advised by Professor Justine Sherry. His research focuses on AI infrastructure and computer systems, with a particular interest in making AI workloads more efficient and observable. His recent work includes FAST, a system for improving the performance of All-to-All GPU communication, and ongoing work on understanding hidden power behaviors and energy waste in GPU clusters.
► TALK 2
DAN VOLZ
Software Engineer, Oracle
— Database Observability
Databases are often treated as standalone systems to be tuned and optimized in isolation. But in modern applications, the database is usually one service within a larger distributed system. A slow query, overloaded session, or internal wait can affect an entire user request, and the root cause is not always visible from either the application or the database alone.
This talk introduces observability as a way to understand systems end to end. We will look at why database visibility matters beyond individual performance tuning, and how metrics, traces, and logs can help connect database behavior to the larger application context.
Dan Volz is a software engineer working at Oracle. I have a Master's and Bachelor's degree from Rice University. My professional and academic interests include a wide range of disciplines ranging across High-Performance Operating Systems, Augmented Reality, Embedded Systems Development, Mobile App Development, and Large-Scale Reliable Distributed Systems.
► TALK 3
CHUNYAN DU
Oracle
— Network Observability
This talk presents Oracle Database Network Monitor (NMON), a network observability framework designed to make database networking measurable, explainable, and debuggable in modern cloud environments. As Oracle Database deployments increasingly span virtual networks, availability domains, load balancers, and distributed services, network behavior has become a critical factor affecting application performance and user experience. Problems such as packet loss, congestion, latency spikes, retransmissions, and routing instability are often transient and difficult to diagnose using traditional monitoring approaches.
NMON addresses these challenges through an end-to-end observability architecture that combines low-overhead telemetry collection, workload-aware correlation, intelligent diagnostics, and future automated remediation. The system leverages both traditional Linux networking interfaces and event-driven eBPF technology to capture real-time flow-level telemetry, including latency, retransmissions, congestion signals, and other network health indicators. This telemetry is streamed to cloud-scale observability platforms where it can be correlated with database operations and customer activities.
Chunyan Du is a software professional at Oracle with deep experience in software architecture, distributed systems, and enterprise-scale engineering. She holds a BE from Beihang University and an MS from UMass Lowell. Her research interests focus on distributed systems, scalable cloud infrastructure, network optimization, AI-driven network diagnostic agents, and reliable software platforms that support high-performance, real-world applications.
Zoom Participation. See announcement.
For More Information:
karenl@andrew.cmu.edu