Skip to content
View unionxic's full-sized avatar

Block or report unionxic

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
unionxic/README.md

Typing SVG

I'm a Software undergraduate at Ajou University, exploring efficient AI systems and reliable GPU communication.

My work focuses on LLM inference, KV cache offloading, and RDMA fault recovery—understanding how data moves across hardware and what happens when communication fails.

Email / LinkedIn

Research

Multi-Tier KV Cache Offloading

  • Developing KV cache supply paths across local DRAM, NVMe, and remote storage using GPUDirect Storage and NVMe-oF over RDMA.
  • Implementing a C++ cuFile backend in vLLM and bandwidth-aware block placement.

RDMA Fault Characterization & Transparent Recovery

  • Validating error propagation across CPU verbs, NCCL, and NVSHMEM through fault injection.
  • Investigating safe replay conditions and prototyping QP-local recovery without application changes.
  • Paper submitted to KSC: Analyzing RDMA Error Conditions and Cross-Layer Error Propagation for Transparent Recovery.

Explore the research repository →

NVIDIA Open-Source Contributions

From reproducing communication failures to proposing upstream fixes.

Project Contribution Links
NVSHMEM Identified incorrect CPU-proxy doorbell-record addressing; fixed internally by NVIDIA. Issue #117
NVSHMEM Reported ignored error completions and submitted a release-build error-handling fix. Issue #120, PR #127
NCCL Reported GDAKI error-propagation gaps and submitted a fix for device waits and persistent host error reporting. Issue #2483, PR #2500

Selected Projects

BudgetOps — Multi-Cloud Cost Management
Developed NCP integration for authentication, cost retrieval, server control, and monitoring. Awarded the Top Prize at Ajou Softcon.
Frontend, Backend

Thermal Imaging-Based Industrial Safety Detection
Ported a camera SDK to ARM-based Raspberry Pi and restored thermal-image collection.

Robust Vehicle Occupant Detection
Built an event-driven Raspberry Pi camera node with automatic process and network recovery.

Toolbox

C/C++ Python Linux CUDA RDMA GDS Docker Git

Education & Awards

Ajou University — B.Eng. in Software
Mar. 2021 – Feb. 2027 (Expected)

  • Top Prize, Ajou Softcon Capstone Design — 2025
  • Encouragement Award, Campus Patent Universiade — 2025
  • Bronze Award, Korean Society for Systems Engineering Invention Competition — 2024

Outside of systems: football, BALLAND, and Arsenal. 🔴⚪

Pinned Loading

  1. YYOAJUNG/budgetops-frontend YYOAJUNG/budgetops-frontend Public

    TypeScript

  2. vllm-gds-kv-offload vllm-gds-kv-offload Public

    vLLM KV cache offload 실험: CPU-staged POSIX vs GPU→NVMe cuFile(GDS) direct path

    Python

  3. rdma-fault-characterization rdma-fault-characterization Public

    Where RDMA faults survive or vanish from the NIC to the application (CPU verbs, NCCL, NCCL GIN, NVSHMEM): pre-registered measurements, five propagation rules, library-internal transparent recovery,…

    Python

  4. KNamuuu/claude-office KNamuuu/claude-office Public

    JavaScript 2 2

  5. NVIDIA/nccl NVIDIA/nccl Public

    Optimized primitives for collective multi-GPU communication

    C++ 5.2k 1.5k

  6. NVIDIA/nvshmem NVIDIA/nvshmem Public

    NVIDIA NVSHMEM is a parallel programming interface for NVIDIA GPUs based on OpenSHMEM. NVSHMEM can significantly reduce multi-process communication and coordination overheads by allowing programmer…

    C++ 596 115