YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Julian Thorne
Systems engineer specializing in low-latency distributed storage, Linux kernel I/O paradigms, and mechanical sympathy in modern hardware.
+-------------------------------------------------------+
| "Make it deterministic, eliminate memory copies, |
| and never guess what the hardware is doing." |
+-------------------------------------------------------+
⚡ Overview
I build low-overhead infrastructure software designed to operate predictably under sustained saturation. Most of my work sits at the boundary between user space and kernel subsystems—leveraging modern Linux primitives like io_uring, eBPF observability hooks, and lock-free concurrency models to extract maximum throughput from commodity NVMe drives and 100GbE networks.
When I am not profiling cache misses or benchmarking storage schedulers, I tinker with mechanical watch movements, pull manual lever espresso shots, and maintain custom lightweight kernels for classic ThinkPads.
🛠️ Technical Focus & Core Stack
All components are built with zero external dependencies where possible and validated under strict memory and latency constraints.
| Domain | Technologies & Tooling | Focus Areas |
|---|---|---|
| Core Systems | Rust, C11, Zig, Go, POSIX C |
Memory safety, zero-cost abstractions, deterministic allocators |
| Kernel & OS | Linux, io_uring, eBPF, XDP, perf |
Asynchronous I/O rings, packet steering, tracepoints, scheduler latency |
| Storage & Data | NVMe-oF, RocksDB, Raft, LSM-Trees |
Write amplification reduction, zero-copy buffer rings, log replication |
| Protocols & IPC | gRPC, Protobuf, FlatBuffers, Unix Domain Sockets |
Cache-friendly binary layout, zero-deserialization access |
| Environment | Neovim, Tmux, Arch Linux, GDB, Valgrind |
Keyboard-driven ergonomics, reproducible offline builds |
🔬 Featured Open-Source Projects
aurora-io
High-throughput, asynchronous user-space I/O runtime engine.
- Pure Rust implementation using modern Linux
io_uringring buffers and fixed buffer registrations. - Achieves sub-microsecond submit-and-reap cycles on PCIe 4.0 NVMe drives without thread context switches.
- Designed for write-heavy append-only telemetry workloads.
knet-trace
Lightweight eBPF probe suite for kernel-level network latency profiling.
- Attaches to kernel socket buffers (
sk_buff) and TCP state transitions to measure queuing delay. - Emits structured histogram statistics directly into terminal viewports with less than 0.8% CPU overhead.
- Real-time threshold monitoring and alerting for TCP retransmissions and window collapse events.
chronolog
Distributed, linearizable commit log backed by strict Raft consensus.
- Implements custom segmented log storage with memory-mapped immutable indexes.
- Optimized for deterministic crash-recovery and microsecond tail-latency guarantees ($P_{99.9} < 450\mu s$).
- Fully pluggable snapshotting architecture with zero background stall on active writes.
nanopack
Zero-allocation binary serialization protocol with SIMD-accelerated framing.
- Direct in-place payload inspection without intermediate heap allocations.
- Portable across x86-64 (AVX2/AVX-512) and ARM64 (NEON).
📈 Engineering Principles
1. Mechanical Sympathy --> Align data layouts with CPU cache lines (64-byte boundaries).
2. Zero Implicit Work --> Avoid hidden allocations, implicit thread pools, and dynamic dispatch.
3. Fail Deterministically--> Crash early, surface explicit errors, and log actionable state.
4. Measure, Don't Guess --> Every optimization requires flame graphs and reproducible traces.
📬 Contact & Coordinates
- Workspace: Munich, Germany (UTC+1)
- PGP Fingerprint:
4E81 B6F2 90D3 7A1C E820 15B9 C3F4 2D88 67A1 E54F - Preferred Communication: Plaintext email, patch submissions via Git format-patch
- Public Keys: Available upon request or via keyserver lookups