Google logo

Google

Google

Full-time · 4 yrs 5 mosFull-time · 4 yrs 5 mos

HybridHybrid

Senior Software Engineer

Senior Software Engineer

Sep 2023 - Present · 1 yr 10 mosSep 2023 to Present · 1 yr 10 mos

United StatesUnited States

• On-prem (airgapped) node & machine management

• OS Image Building, Installation & Upgrade

• Ubuntu, Rocky Linux OS

• K8S

• GPU: Dealing with H200 B200 B300 w/ 570/580 Nvidia Driver; Perf Boost.

Google Distributed Cloud - Hosted (GDC-H)

TL;DR: K8s cluster node lifecycle management for air-gapped cloud infra.

L3->L5 in 2.5y.

• Airgapped Solution for On-Prem Cloud Infra development (0->1)

• Cluster Node Lifecycle Management

• Baremetal Machine & OS Management (Ubuntu, Rocky CentOS Build, Installation and Upgrade)

• System Manager, AddOn Manager, Multi-Tenancy

• Kubernetes Cluster, Control Plane (Kubelet, Containerd, etcd)

• Cloud Native Service Mesh Open Source Project: Envoy-proxy(C++), Istio(Golang)

• On-Demand Filter Chain - Service Mesh Proxy listener Discovery Service, eliminated time waiting for inactive dependencies

• Breaking changes to the envoy proxy thread mode. Nonblocking worker-master-worker thread locks.

• Designed and implemented target-aware callback watchers - brought quick visibility for blocker dependencies/Certs of Envoy init-managers.

• Enabled config dumping for unready targets of blocked listeners during warming process with Protobuf message.

• Provided early-starting strategies for Envoy listeners with placeholder components(filter chains) - avoided blockers and decreased response latency.

• Enabled dynamic listener filter-chain dependency update on new connection requests - automatic connection retrial callbacks - worker-master-worker thread communication.

• More Istio APIs - improved the scalability at large mesh and reduced the usage of resources per instance.

• Researcher and Software Developer

• AI Infra: Detection and Tracking System for experiment mini cars

• SenseStudy platform development

• Web Game Development in AI Education

• Tools Development for data labeling, model training, system visualization

• Image Semantic Segmentation, Instance Segmentation(Mask R-CNN, Deeplabv3)

• Object affinity relationship with depth information(ResNet backboned)

• Model Compression and Acceleration for Deep Neural Networks

• Video Segmentation with Optical Flow(Traditional + FlowNet)

• Image Inpainting to replace passersby with background information(Gated Conv + Spectral Norm)

• Made Docker Images for deep learning projects and environments