//assets.cakeresume.com/assets/contentarea/contentarea-2d64eca8dae2a1ffb31e0fb1df887bbad072c7c19734d7d30fa14884bbcc299a.css

Voda scheduler is a GPU scheduler for elastic/distributed deep learning workloads based on Kubernetes, Kubeflow Training Operator and Horovod.

Features:
 - Rich Scheduling Algorithms (with Resource Elasticity)
 - Topology-Aware Scheduling & Worker Migration
 - Microservices architecture
 - Node addition/deletion awareness
 - Heterogeneous scheduling
 - Fault-Tolerance

https://github.com/heyfey/vodascheduler

Voda Scheduler

Senior Software Engineer

謝宗佐

Work on Compiler & ML systems & Kubernetes
Skilled in parallel computing, distributed systems
Code in C++, C, Golang, python

SCOPE Lab (Scientific Computing On Parallel Environment)
- Parallel Computing
- Distributed Systems
- ML system design


國立清華大學

Synopsys

VCS - verilog compile simulator

(C++, C, Compiler Optimization, Python)

• Developed and supported debug functionality for SystemVerilog compile simulator
• Parallelized design traversal in FSDB dumping, speeding up runtime by 3x with multithreading
• Optimized FSDB gate-data loading and reduced 10-60% runtime memory footprint, benefiting customers running hundreds-of-GB designs with tight memory budgets
• Initiated a project to migrate STL hash tables to Abseil Swiss tables for a millions-line code base, reducing up to 70% peak memory usage and speeding up hash table queries by 100%
• Debugged and resolved urgent customer issues in tight timelines (< 1 day TAT) with limited information, tools, and access to the environment

Voda Scheduler: GPU scheduler for elastic/distributed deep learning workloads in Kubernetes cluster
https://github.com/heyfey/vodascheduler (30+ stars)

(Golang, Python, Distributed Systems)

• Designed and built a GPU scheduler on top of Kubernetes and several open-source projects using microservices architecture
• Sped up overall training time by 2.38x and increased cluster utilization by 1.4x with
        -- State-of-the-art scheduling algorithms with resource elasticity
        -- Topology-aware scheduling & worker migration to consolidate resources
• Designed and implemented the heterogeneous scheduling, auto-scaling, and fault-tolerant mechanisms for the system

Research Assistant

(C, Python, Machine Learning)

• Designed algorithm & firmware for real-time embedded system
• Introduced machine learning and data visualization to automate issue analysis, saving 90% of human effort

Share