The address book every process carries

Processes Threads and the Kernel · Open board

The kernel in the middle, handing every program a private-looking stretch of memory. This lesson opens the book the kernel keeps for each program, and asks what happens when the real memory behind it runs short.

Before you start. RAM is fast and small, the disk is huge and far slower, and a program runs well only when what it needs sits in the fast part (The Memory Hierarchy (cache, locality, the memory wall)). The address a program uses is not where its data really is: the kernel, the one program with full control of the hardware, translates every address on the way (Processes, Threads & the Kernel). And the kernel moves memory around in fixed-size pieces called pages, the same pieces it uses to cache files (The drawer that says it saved).

The idea in 30 seconds

A program’s memory is cut into fixed-size pages, and the kernel keeps a page table for each program that says where each page really is: in RAM, on disk, or nowhere yet.
Asking for memory only writes a promise into that table; real memory is handed out one page at a time, when the program first touches it.
That first touch causes a page fault: the processor stops the program, the kernel fills in the page, and the program carries on without knowing.
Every program has a working set, the pages it has used recently; given room for it, a program rarely faults, and given less, it thrashes, spending its time waiting for the disk.
When a program copies itself (fork), the original and the copy share every page until one of them writes, and only the written pages are ever copied (copy-on-write).

Read the story · Skip to the numbers