📖 Tutorial: Systems in the Age of AI (SysAI)
🎯 Tutorial Overview
AI workloads are rapidly reshaping system design assumptions, exposing limitations in resource management, data paths, distributed execution, and observability that were not built for large-scale, heterogeneous, and latency-sensitive AI environments. This tutorial highlights cross-layer design across kernel, runtime, and distributed layers to support AI-native workloads.
The tutorial is structured around two complementary tracks: System Operations and Optimization Acceleration in the AI Era and System and Distributed Scheduling Solutions for AI Workloads. The former focuses on heterogeneous resource management, distributed runtimes, and container-oriented data-path optimization to improve efficiency and scalability of AI execution. The latter explores how AI enhances system intelligence through observability, agent-aware scheduling, and large-model integration into system development and operational workflows.
📘 System Operations and Optimization Acceleration in the AI Era
TrIO - A Flattened and Cooperative I/O Mechanism for Container Environments
📅 Time: 09:00 - 09:40 AM
Topic Abstract:
This tutorial delves into the I/O bottleneck of container image loading and introduces a novel solution, TrIO. TrIO proposes a memory-oriented image abstraction and a runtime page cache to achieve efficient image service. TrIO is open sourced and integrated into openEuler, and published in FAST'25.
Yubo Liu
Research scientist at Huawei. Focuses on high-performance storage, memory, and HPC systems. Published in FAST and TOS. PC member for MSST, IPDPS, NAS.
🧩 Mystery Gift Piece
EulerCopilot - Key Technologies and Practices of Intelligent Fault Diagnosis and Optimization in openEuler
📅 Time: 09:40 - 10:20 AM
Topic Abstract:
This tutorial introduces openEuler's intelligent O&M architecture for cloud and AI infrastructure. It focuses on key technologies for intelligent fault diagnosis and performance optimization, including multi-Agent collaboration for precise root cause analysis, multi-source data integration for adaptive system optimization, and experience-driven Agent skill evolution.
Wang Lei
Chief architect of EulerCopilot at Huawei. Expert in LLM Agents, Serverless, and DevOps. Author of "Microservice Architecture and Practice" and other books.
🧩 Mystery Gift Piece
📘 System and Distributed Scheduling Solutions for AI Workloads
LLM-LA - LLM Load-Aware Load Balancing
📅 Time: 11:00 - 11:40 AM
Topic Abstract:
Variability in LLM input and output token lengths leads to load imbalance in multi-instance serving systems, where short and long requests share queues and increase waiting time and tail latency. This tutorial introduces LLM-LA, a load-aware request management system for LLM workloads, explaining its design and practical deployment in serving frameworks.
Saeid Ghafouri
Research Scientist at Huawei Edinburgh. PhD from Queen Mary University of London. Focuses on LLM serving infrastructure and distributed inference systems.
🧩 Mystery Gift Piece
openYuanrong - Distributed Serverless AI Computing Engine
📅 Time: 11:40 AM - 12:20 PM
Topic Abstract:
This tutorial introduces openYuanrong, a Serverless distributed compute engine that unifies diverse applications, from AI and big data to microservices, on a single, streamlined architecture. It provides multi-language function interfaces that simplify the development of complex distributed applications. Powered by dynamic scheduling and efficient data sharing, openYuanrong ensures high-performance execution and maximum cluster resource utilization.
Mohamed Kassem
Principal engineer at Huawei Edinburgh. PhD from University of Edinburgh (2020). Focuses on LLM serving engines and Agentic AI. Former CTO of a multi-million tech company.
🧩 Mystery Gift Piece
🎁 Collect All Puzzle Pieces & Win a Prize!
Attend all tutorials, collect every puzzle piece, and bring them to the registration desk to claim your reward.
