gsplpg/johyojun339-lgtm

KR 한국어	US English	JP 日本語	CN 简体中文	ES Español
FR Français	DE Deutsch	RU Русский	SA العربية	IN हिन्दी

---

Hyojun Jo (조효준)

Systems Software Engineer focusing on Linux Kernel Observability, High-Throughput Packet Processing, and Low-Latency Network Infrastructure.

I design and optimize low-level software where microseconds matter. My work revolves around eBPF kernel instrumentation, XDP-driven network pipelines, memory-efficient data structures, and deterministic telemetry for distributed production clusters.

• 📍 Location: Seoul, South Korea (UTC+9)
• 💼 Primary Focus: Linux Kernel Subsystems, eBPF/XDP, Network Performance Engineering
• 💬 Ask me about: Cache locality, zero-copy socket buffers, ring buffers, eBPF verifier quirks, memory-mapped I/O
• 📬 Contact: johyojun339@gmail.com

---

🛠️ Technical Arsenal

Languages & Core Runtimes

Layer	Technologies & Tools
Primary Systems	Rust (nightly/stable), C (C11/C23), Go (1.23+), x86-64 / AArch64 Assembly
Kernel & Tracing	Linux eBPF (libbpf, aya-rs), XDP, kprobes/uprobes, tracepoints, perf_events
Scripting & Automation	POSIX Shell / Bash, Python 3, PowerShell Core, AWK / Sed

Operating Systems & Infrastructure

Domain	Stack & Environment
Kernel Subsystems	Linux VFS, Netfilter / nftables, Socket Buffers (sk_buff), cgroups v2, tc-bpf
Observability & Metrics	OpenTelemetry, Prometheus, DuckDB, ClickHouse, Grafana, Vector
Tooling & Profiling	bpftool, perf, gdb, valgrind, strace, flamegraph, LLVM / Clang 18+
Networking Protocols	TCP/IP, UDP, QUIC / HTTP3, WireGuard, BGP / EVPN fundamentals

---

🚀 Active Open Source Projects

+--------------------------------------------------------------------------+
|  flow-kprobe  [Rust / C / eBPF]                                          |
|  Zero-overhead Linux kernel probe engine for TCP socket life-cycle       |
|  tracking and microsecond packet latency attribution.                    |
+--------------------------------------------------------------------------+
|  wire-sieve   [C / XDP]                                                  |
|  High-throughput L4 stateless packet dispatcher running entirely at the |
|  NIC driver level, capable of handling 14.8M pps per 10GbE core.         |
+--------------------------------------------------------------------------+
|  nanotrace    [Go / Rust / libbpf]                                       |
|  Lightweight CLI utility to inspect tail latencies across Linux system   |
|  calls with sub-microsecond histogram visualizers in the terminal.       |
+--------------------------------------------------------------------------+

1. flow-kprobe

• Core Architecture: Hooks into tcp_v4_connect, tcp_set_state, and tcp_close via modern CO-RE (Compile Once - Run Everywhere) eBPF programs.
• Key Feature: Streams low-overhead ring buffer events to user-space without context-switching churn.
• Benchmark: Under a 100,000 req/sec benchmark load, CPU utilization stays below 0.8% on a standard 8-core host.

2. wire-sieve

• Core Architecture: Bypasses the Linux network stack using XDP native mode (XDP_TX / XDP_PASS) to drop malformed traffic and re-route validated packets directly at the network interface layer.
• Design Philosophy: Strict mechanical sympathy, zero-allocation memory pools, cache-line-aligned packet metadata structures.

3. nanotrace

• Core Architecture: Attaches to sys_enter and sys_exit tracepoints to measure exact system call duration across storage and network threads.
• Output: Terminal-based ASCII percentile ladders (p50, p95, p99, p99.9) rendered without external web dependencies.

---

📊 Engineering Principles & Code Aesthetics

1. Mechanical Sympathy Above All: Software runs on hardware. Cache-line alignment, branch predictor ergonomics, and NUMA node locality take precedence over theoretical abstractions.
2. Explicit Over Implicit: Zero-magic codebases. Explicit error returns, strict ownership semantics, and minimal runtime indirection.
3. Observability is Not an Afterthought: If a system cannot be inspected under load with zero performance penalty, it is incomplete.
4. Data-Oriented Design: Optimize for contiguous memory arrays and vectorized operations over deep object hierarchies.

---

📈 System Benchmark Snapshot

Metric	Baseline (Stock Linux TCP)	Tuned eBPF / XDP Bypass	Improvement
L4 Packet Filtering	1.82 Mpps (Single Core)	14.45 Mpps (Single Core)	+693.9%
TCP State Tracing Overhead	4.2% Host CPU	0.4% Host CPU	-90.4%
System Call Tail Latency (p99)	142.6 microseconds	18.2 microseconds	-87.2%
Telemetry Buffer Memory Footprint	64 MB Heap Buffers	4 MB Ring Buffer	-93.7%

---

📬 Get In Touch

• GitHub Issues & Discussions: Feel free to open an issue or start a thread in any of my public repositories.
• Email: johyojun339@gmail.com
• GPG Key: 4A8F B120 9C7E 38A1 (Signed commits only)
• Time Zone: KST (UTC+9), usually active during daytime and late-night coding sessions.