A walkthrough of OSTEP Chapter 13 covering how modern operating systems abstract physical memory through address spaces. Explains the historical evolution from single-process systems to multiprogramming and time sharing, the three components of an address space (code, heap, stack), and the OS goals of transparency, efficiency, and protection in virtual memory. Also extends the discussion to modern GPU memory challenges, including the memory wall problem in LLM inference, HBM, CUDA Unified Memory, KV cache offloading, and CXL-attached memory tiers.

4m read timeFrom muratbuffalo.blogspot.com
Post cover image
Table of contents
Multiprogramming and Time SharingThe Address Space and VirtualizationGPUs and Beyond

Sort: