thrashing

Thrashing is what happens when a machine is so short of RAM that it spends almost all its time shuffling pages between memory and disk and almost none actually computing. The work the programs want to do barely advances while the disk light stays on solid; the system feels frozen even though the CPU is technically busy. It is the failure mode at the far end of swapping.

Here is the vicious cycle, step by step. The combined working set of the running programs — the pages they each need soon — is larger than physical RAM. To free a frame for one program, the OS must evict a page another program will need again almost immediately, which then faults that page back in by evicting yet another still-needed page. Every program keeps faulting on pages that were just thrown out, each fault a slow major fault waiting on the disk. Useful CPU work stalls behind a flood of disk reads, so throughput collapses toward zero even though utilisation counters may look high.

Thrashing is not a gentle slowdown, it is a cliff: a little memory pressure is handled smoothly, but once the hot working set exceeds RAM the system can fall off into near-total paralysis. The cures are about the working set, not the swap: run fewer or smaller programs so the hot pages fit, add real RAM, or improve locality so each program needs fewer pages resident at once. Adding more swap space does not help and often makes it worse, because it lets the overcommit get deeper before collapsing.