16 / 22 Virtual Memory & Paging

Four Percent Busy

A box with 8 GB of RAM and 32 GB of swap runs four processes, each with a working set of about 3 GB. Monitoring shows:

metric before now
major page faults / s 20 200,000
CPU utilization 78% 4%
swap device utilization 3% 100%

Adding a fifth identical process makes every number worse. What is going on, and what actually helps?

A

The replacement policy is FIFO rather than LRU; switching to LRU removes the faults

B

The combined working sets (12 GB) exceed physical memory, so pages are evicted before they are reused; the fix is to run fewer processes at once, not to add swap

C

Swap is undersized; growing the swap file lets the working sets fit and restores throughput

D

The CPU is starved, so raising the scheduling priority of the faulting processes lets them make progress