How Many Frames per Process?

Each process needs a minimum number of frames so its instruction can complete after a fault. The OS chooses among:

• Equal allocation — same frames per process.
• Proportional allocation — frames proportional to process size or priority.

Global vs Local Replacement

• Local replacement — process replaces among its own frames.
• Global replacement — process can take any frame in the system. Higher throughput on average; but a misbehaving process can hog memory.

What is Thrashing?

A process is thrashing when it spends more time paging than executing. It happens when the working set does not fit in the allocated frames — every reference is a fault.

CPU utilization shows a characteristic curve: rises with multiprogramming up to a knee, then collapses as thrashing sets in.

Causes

• Too many processes for available RAM.
• A long-running process whose working set keeps shifting.
• Bad replacement decisions (FIFO, MFU under wrong workload).

Combating Thrashing

1. Working-Set Model (Denning) — for each process, the set of pages referenced in the last Δ time units. Sum of working-set sizes ≤ total frames; if not, suspend a process.
2. Page-Fault Frequency (PFF) — keep each process's fault rate within acceptable bounds. If a process's PFF is too high, give it more frames; if too low, take some away.
3. Reduce multiprogramming level — swap out an entire process.
4. Add RAM — the brute-force solution.

Working-Set Computation Example

Reference string ... 2 6 1 5 7 7 7 7 5 1 6 2 3 4 1 2 3 4 4 4 ...

With Δ = 10, working set at time t = unique pages in last 10 references.

Other Memory Pressures

• Heap fragmentation — long-running servers may need allocator tuning.
• TLB churn — if the working set has too many distinct pages, even hot data suffers TLB misses.
• NUMA effects — on multi-socket boxes, accessing remote-node memory is slower.

Summary

• Thrashing collapses throughput and feeds itself unless the OS intervenes.
• Working-set and PFF heuristics decide frame allocation.
• The cure is more memory, fewer processes, or smarter replacement.