Measurement-driven systems performance.

For people who read the disassembly. Storage, memory, kernel, compute — and the FPGAs that beat them.

→ weekly measurement posts → monthly deep dives → quarterly interactives

┌─[ featured ]── 2026-04-11 · 3m read · cache · cpu · memory ┐

A cache-line visualizer

An interactive that lets you feel cache-line behavior: array-of-structs vs struct-of-arrays, false sharing in action, 64-byte alignment, the padding trick — the visceral version of things every systems engineer has memorized.

# cache # false-sharing # visualization # interactive

read post →

└ ┘

┌─[ recent posts ]─ ┐

date

layer

title

topics

read

depth

2026-04-11

app · syscall

io_uring vs epoll vs blocking read on Linux 6.18 LTS — a fair benchmark

#io_uring #epoll #linux

2026-04-11

app · cache

What HNSW actually does to your cache

#hnsw #vector-search #cache

└ ┘

┌─[ by topic ]─ ┐

storage

├─ io_uring
├─ NVMe Gen5
├─ SPDK
├─ kernel bypass
└─ GPUDirect storage

memory & cache

├─ NUMA
├─ L1 / L2 / L3
├─ huge pages
├─ false sharing
└─ CXL tiering

cpu & uarch

├─ pipelines
├─ branch prediction
├─ SIMD / AVX
├─ FPGA / RTL
└─ microcode

concurrency

├─ lock-free
├─ ring buffers
├─ seqlocks
├─ RCU
└─ epoch GC

numerical

├─ BLAS internals
├─ vectorization
├─ AoS / SoA
└─ intrinsics

kernel & sched

├─ sched_ext
├─ eBPF
├─ syscall overhead
└─ real-time tuning

└ ┘