Memory pressure

Lots of free memory → just pull a page off the free list and assign it to the faulting page.

Low memory → memory pressure forces the OS to page out actively-unused pages to make room. That's eviction, driven by page replacement policies.

Memory as a cache

Main memory holds a subset of virtual pages — it's a cache for them. Goal: minimize misses, maximize hits.

Average memory access time:

AMAT = (P_hit × T_m) + (P_miss × T_d)

• T_m — cost of accessing memory
• T_d — cost of accessing disk

Miss types

Miss	Why
Compulsory	Cache starts empty
Capacity	Cache ran out of space
Conflict	Limits on where an item can sit (e.g. two items hash to the same spot)

Replacement policies

Optimal (Belady's): replace the page accessed furthest in the future. Perfect — and impossible to know for sure. Use it as a baseline; look at the past, can't know the future.

Policy	Idea
FIFO	First in, first out — arguably worse than optimal
Random	Luck of the draw — theoretically can match optimal
LRU	Least recently used — frequency + recency; less likely to kick out a page you just needed
MRU	Most recently used — bad; ignores locality

Next up: analyze complex workloads and chart how each policy does.