'Thrashing' — near-total performance collapse under multiprogramming — was solved by Denning's 1968 working-set model

Once early operating systems began multiprogramming — running several jobs at once to share the same scarce physical memory beneath a virtual-memory layer — a new failure mode appeared: thrashing, which Peter J. Denning describes as "a condition of near-total performance collapse," colloquially "paging to death." As competing jobs each churn pages in and out of physical memory faster than useful work can proceed, the system spends nearly all its time servicing page faults rather than computing.

The fix was Denning's own 1968 contribution: the working-set model, which tracks each process's recently referenced pages (its "working set") and only schedules a process for execution when its full working set can be held in physical memory simultaneously — preventing over-multiprogramming from starving every job of the memory it actually needs in the near term.

This is the specific 1960s mechanism — not just virtual memory generally, but its failure mode and named fix — that reappears in modern LLM-serving systems under memory pressure. See claim-llm-serving-vllm-reimports-os-preemption-and-thrashing for the vocabulary's return, and claim-atlas-1959-first-virtual-memory-one-level-store for the virtual-memory system this failure mode first appeared in.