🗂 Page Replacement Policies

About this simulation

When a running process references a page that isn't currently in physical memory, the operating system takes a page fault and must load it — evicting another resident page if all frames are full. Which page gets evicted is decided by a page-replacement policy. This simulation runs the same request sequence through four classic policies at once, one row of memory frames per policy, so you can watch their eviction decisions and fault counts diverge in real time.

The four policies

• FIFO — evicts whichever resident page was loaded longest ago, regardless of how recently it was used. Simple, but can suffer Belady's anomaly: adding more frames can sometimes increase faults.
• LRU — evicts the page that hasn't been touched for the longest time, on the theory that recent use predicts near-future use. Needs per-access bookkeeping to implement exactly.
• Clock (second-chance) — a cheap approximation of LRU. Frames sit in a circular list with a reference bit; a "hand" sweeps around clearing bits and only evicts a page whose bit is already 0, giving recently-touched pages a second chance.
• Optimal (Belady's MIN) — evicts whichever resident page will not be needed for the longest time in the future. It requires knowing the whole request sequence in advance, so it is never implementable online — but it sets the theoretical lower bound on fault count that real policies are measured against.

Controls

• Frame count — how many memory frames each policy has to work with (3–6). Fewer frames means more faults.
• Feed speed — how fast requests are fed into the four rows.
• Request pattern — the shape of the reference string: a tight working set with locality, a repeating loop, a Belady worst-case sequence, or uniform random.
• ↺ New sequence — generates a fresh reference string and resets every row's frames and fault counters.

How to read the scene

Each row of boxes is one policy's memory frames; the number on a box is the resident page. A box flashes green on a hit and orange on a fault when a new page is loaded. The vertical bar at the end of each row grows with that policy's running fault count, so the row that grows tallest is currently faulting the most.