Page Tables Are Not Free: The 1/512 Rule That Breaks Under Shared Mappings
A 4 KiB data page costs 8 bytes to map. That is a ratio of 1 to 512. When one address space maps a page once, the page table overhead is roughly one 512th of the memory it describes, which is why most
Responses
Markdown supported · ⌘↵ to post
Ahmet Özel
AI Engineer. Computer Vision, RAG and LLM agents.
The shared mapping examples make the address-space multiplier much clearer than the usual PTE-to-page ratio. One distinction I would make explicit in the checklist is processes versus threads: threads sharing an address space do not each introduce another complete set of page tables for that mapping.
A useful companion measurement would compare a threaded worker pool with separate processes that touch the same shared region. Recording VmPTE once per address space, alongside resident data and which pages were actually touched, would help distinguish duplicated translation structures from a large virtual reservation that has not populated them yet.