M.S. Student in Software Engineering · Joint Training @ National University of Defense Technology (NUDT)

Research focus: workflow scheduling algorithms based on Deep Reinforcement Learning

📍 Changsha, China

• 🎓 M.S. student in Software Engineering at Changsha University of Science & Technology（CSUST）, jointly trained at the National University of Defense Technology (NUDT), graduating in 2027.
• 🧠 Researching workflow scheduling algorithms based on Deep Reinforcement Learning
• 🛠️ AI full-stack engineer working on Agent application development — from LLM orchestration to production-grade backend services.
• 🚀 Goal: becoming a software engineer at a leading Internet technology company.

A high-fidelity discrete-event simulator for scientific workflow scheduling research, built on the WorkflowSim / CloudSim codebase.

Instead of the usual "fork and hope it runs" simulator maintenance, this project provides:

• ✅ Strict input validation for DAX 2.1 XML, WfCommons WfFormat JSON, and WfInstances 1.5 JSON
• 🧪 Explicit experiment configuration — every parameter declared in code, reproducible from a single right-click in IntelliJ IDEA
• 📦 Auditable evidence artifacts — each run emits a verifiable manifest.json / metrics.json / events.jsonl bundle
• 🔬 Semantically tested algorithm tracks — online ready-Job scheduling (FCFS, MINMIN, MAXMIN, MCT, ...), static DAG mapping (HEFT, CPOP, ETF, DLS, PEFT), and an RL scheduling track (RlEnvironment + RL_POLICY)

The RL track exposes a state/action/reward environment contract with a deterministic closed loop, enabling external DRL trainers to be bridged out-of-process — the foundation of my research on DRL-based workflow scheduling.

Also includes Fat-tree topology-aware link-contention modeling (fatTreeContentionV1) and a 360-run deterministic Fat-tree × scheduling campaign, demonstrating how network contention shifts per-DAG rankings even when average rankings hold.

Highlights: 446 tests passing under mvn verify · full HEFT/CPOP paper reproduction (Topcuoglu et al., IEEE TPDS 2002) · JDK 17 + Maven