COMP 620   Computer Systems Seminar

Instructor: Jiarong Xing, Office: DCH 2099
Lectures: 3:00-4:15 pm, Friday
Location: DCH 1042


Course Description

This graduate seminar explores the design and implementation of modern computer systems. We will read and discuss classic and contemporary research papers across various system topics, with an emphasis on critical analysis and in-class discussion.

Topics include (but are not limited to):

Format

The class meets once per week. Each week, there will be a leader of the discussion who will present the paper and lead the discussion. The rest of the class will deeply engage in the discussion, asking questions, providing feedback, and sharing their own insights.

After each class, students will need to submit a summary of the discussion.

Depending on the registered students, we might invite external speakers to give talks on related topics.

There is no exam for this course.

Grading

Accommodations

Students with a documented disability needing academic adjustments or accommodations are encouraged to contact the instructor and Disability Support Services (Allen Center, Room 111).

Tentative Schedule

Date Topic Paper/Talk Speaker
1/16/2026IntroductionCourse Logistics and New Trends in Computer SystemsJiarong Xing
1/23/2026GPU SharingEfficient Performance-Aware GPU Sharing with Compatibility and Isolation through Kernel Space InterceptionRixin Liu
1/30/2026Prompt EvolutionGEPA: Reflective Prompt Evolution Can Outperform Reinforcement LearningLouie Lu
2/6/2026Research methodologyYou and your researchJiarong Xing
SPRING RECESS (NO SCHEDULED CLASSES)
2/20/2026GPU SharingMLaaS in the Wild: Workload Analysis and Scheduling in Large-Scale Heterogeneous GPU ClustersXingqi Cui
2/27/2026LLM inferenceMEDUSA: Simple LLM Inference Acceleration Framework with Multiple Decoding HeadsShen Zhang
3/6/2026GPU OSMotivation and The Future of GPU OSJiarong Xing
3/13/2026MoE serving optimizationExternal lecture: Resource-Efficient MoE LLM Serving via Fine-Grained Expert OffloadingHanfei Yu
SPRING BREAK (NO SCHEDULED CLASSES)
3/27/2026LLM RoutingLookahead Routing for Large Language ModelsGeorge Zhang
4/3/2026ADRSExternal lecture: AI-Driven Discovery: From Algorithm Generation to Self-Improving Research LoopsShu Liu
4/10/2026AI for kernel generationExternal lecture: LEO: Cross-Vendor GPU Performance Root Cause Analysis and LLM-Guided OptimizationYuning Xia
4/17/2026AI training privacyKeeping LLMs from Exposing Sensitive Data - A VaultGemma Case StudyPeter Pham
4/24/2026AI training resiliencyResilient Distributed Training under Failures: A ReCycle Case StudyChaoHsuan Ho