How Linux Reclaims Memory: Inside the Page Frame Reclaim Algorithm

Page Frame Reclaim Algorithm (PFRA)

Virtual memory is managed in pages, typically 4 KB, or 2 MB when using huge pages. PFRA reclaims memory based on page types.

1. Page Types

Non‑reclaimable pages: locked pages, kernel pages, reserved pages.

Swappable pages: anonymous memory pages.

Sync pages: pages that have a corresponding disk file.

Discardable pages: static or discarded pages.

All types except the first can be reclaimed by PFRA.

2. PFRA Mechanisms

kswapd kernel thread.

pdflush kernel thread.

3. kswapd

The kswapd daemon ensures enough free memory is available. When free memory falls below pages_low, kswapd releases pages in batches of 32 until free memory reaches pages_high. It uses a double‑LRU list: unmodified pages are moved to the free list; modified pages with a backing file are written back to disk; modified anonymous pages are written to swap.

4. pdflush

pdflush writes dirty pages belonging to files back to disk. When about 10 % of pages become dirty, pdflush starts syncing them; the threshold can be tuned via vm.dirty_background_ratio.

5. vmstat‑related Parameters

Key vmstat parameters that influence memory reclamation include pages_low, pages_high, and vm.dirty_background_ratio.

6. Example

An analysis of an I/O spike shows massive virtual‑memory usage: many disk blocks are mapped to pages, free memory stays around 17 MB, kswapd pulls memory from buffers to keep the free list, and dirty pages are written to swap.

7. Summary

Page faults occur.

Heavy swap usage indicates memory shortage.