Blog — Kiran Hombal

What shared CXL memory is, why hardware coherence covers only a sliver of it, and how to share data on it anyway: a general-audience walk from the memory wall to our lab's Megalon (OSDI '26) and Prism (SOSP '26).

A visual reading of the first production-scale characterization of an AI coding agent: sampled GitHub Copilot traces covering 13.5M sessions, 3.2M users and 760.5M LLM calls.