Hsu-Tzu Ting

Hello! I am an incoming PhD student in Computer Science at University of Illinois Urbana-Champaign. Last year, I completed a summer internship at Google, where I gained hands-on experience in cutting-edge technology. My research focuses on cloud computing and distributed systems, and I’m thrilled to share that my paper, KubeComp, was accepted at UCC!

During my undergraduate years, I participated twice in international student cluster competitions. Our team won the championship at SC22 in Dallas, United States, and placed second at ISC23 in Hamburg, Germany. Preparing for these competitions sparked my interest in high-performance computing (HPC).

In 2023, I spent a semester as an exchange student at TU Dresden in Germany, where I enjoyed the enriching experience of immersing myself in a different culture.

news

Jun 13, 2025	I have participated in the ISC Student Volunteer Program at Hamburg, Germany!
Feb 21, 2025	I am admitted to the Computer Science PhD program at the University of Illinois Urbana-Champaign!
Feb 21, 2025	“PCIe Bandwidth-Aware Scheduling for Multi-Instance GPU” received the Best Paper Award at HPCAsia!
Oct 26, 2024	KubeComp was accepted to UCC 2024!
Sep 14, 2024	Completed a Software Engineering internship at Google this summer!

selected publications

UCC
KubeComp: A Resource-Centric Composable Container Orchestrator for GPU Pooling

Hsu-Tzu Ting, Chou Jerry, Ming-Hung Chen, and 2 more authors

IEEE/ACM 17th International Conference on Utility and Cloud Computing, 2024

Abs DOI Bib PDF Code

Composable infrastructure, designed to allocate hardware from a disaggregated resource pool dynamically, aligns seamlessly with the resource-centric nature of cloud computing. Although a few GPU disaggregation systems for reconfigurable bare metal devices exist, the corresponding software stacks are lagging. In this work, we designed and implemented KubeComp, the first solution for Kubernetes to support GPU pooling based on a composable infrastructure. KubeComp makes Kubernetes aware of the underlying composable infrastructure and provides on-demand, optimized, and automated resource management for service deployment. With the support of KubeComp, resource utilization is boosted, and the job wait time is reduced by up to 89% for running jobs with diverse CPU-GPU ratios.
@article{kubecomp, author = {Ting, Hsu-Tzu and Jerry, Chou and Chen, Ming-Hung and Chung, I-Hsin and Pan, Huaiyang}, journal = {IEEE/ACM 17th International Conference on Utility and Cloud Computing}, title = {KubeComp: A Resource-Centric Composable Container Orchestrator for GPU Pooling}, year = {2024}, keywords = {Composable, GPU pooling, Container, Scheduling, Cloud Computing}, doi = {10.1109/UCC63386.2024.00014}, }