3 articles
From L1 to main memory and beyond — understanding why your 'fast' code might be waiting on data it already asked for.
A deep dive into the Linux I/O stack — from userspace syscalls through the VFS layer, block I/O schedulers, and down to the hardware. Covers io_uring, buffered vs direct I/O, and practical tuning strategies.
Practical techniques for managing what goes into your LLM's context window and why it matters.