CPS 346 Lecture notes: Allocation and Thrashing

Coverage: [OSCJ] §§9.5-9.6 (pp. 418-427)

Overview

When should we run the page replacement algorithm?

Two main questions in the study of virtual memory:

maintain 3 free frames at all times
consider a machine where all memory-reference instructions have only 1 memory address → need 2 frames of memory
- now consider indirect modes of addressing
- potentially every page in virtual memory could be touched, and the entire virtual memory must be in physical memory
- place a limit the levels of indirection
the minimum number of frames per process is defined by the computer architecture
the maximum number of frames is defined by the amount of available physical memory

m frames, n processes
equal allocation: give each process m/n frames
alternative is to recognize that various processes will need different amounts of memory
- consider
  - 1k frame size
  - 62 free frames
  - student process requiring: 10k
  - interactive database requiring: 127k
  it makes no sense to give each process 31 frames
  
  the student process needs no more than 10 so the additional 21 frames are wasted
proportional allocation:
- m frames available
- size of virtual memory for process p_i is s_i
- S = Sigma s_i
- a_i = (s_i/S) * m
student process gets 4 frames = (10/137)*62

database gets 57 frames = (127/137)*62
what about priority? use priority rather than relative size to determine ratio of frames

is page-replacement global or local?
global: one process can select a replacement frame from the set of all frames (i.e., one process can take a frame from another or itself) (e.g., high priority processes can take the frames of low priority processes)
local: each process can only select from its own set of allocated frames
local page replacement is more predictable;
depends on no external factors
a process which uses global page replacement cannot predict the page fault rate;
may execute in 0.5 seconds once and 10.3 on another run
overall, global replacement results in greater system throughput

simple thrashing: 1 process of 2 pages only allocated 1 frame
high page activity is called thrashing
a process is thrashing if it spends more time paging than executing
scenario:
The process scheduler see that CPU utilization is low.

So we increase the degree of multiprogramming by introducing a new process into the system.

One process now needs more frames.

It starts faulting and takes away frames from other processes. (i.e., global-page replacement).

These processes need those pages and thus they start to fault, taking frames from other processes.

These faulting processes must use the paging device to swap pages in and out.

As they queue up on the paging device, the ready-queue empties.

However, as processes wait on the paging device, CPU utilization decreases.

The process scheduler sees the decreasing CPU utilization and increases the degree of multiprogramming, ad infinitum.

The result is thrashing, page-fault rates increase tremendously, effective memory access time increases, no work is getting done, because all the processes are spending their time paging.
Fig. 9.18
we can limit the effects of thrashing by using a local replacement algorithm (or priority replacement algorithm)
to really prevent thrashing, we must provide processes with as many frames as they need.

as a process executes, it moves from locality to locality
a locality is a set of pages used together. A program is generally composed of several localities which may overlap.
for instance, when a function is called, it defines a new locality
the locality model is the basic unstated assumption behind cache; if accesses to any types of data were random rather than patterned, cache would be useless.
if we allocate enough for a process to accommodate its current locality, it will fault for the pages in its locality until all of these pages are in memory. Then it will not fault again until it changes locality.
if we allocate fewer frames than the size of the current locality, the process will thrash since it cannot keep in memory all the pages which it is actively uses.

based on the assumption of locality
Δ defines the working-set window: some # of memory references
examine the most recent Δ page references
the set of pages in the most recent Δ is the working set or an approximation of the program's locality.
Fig. 9.20
the accuracy of the working set depends on the selection of Δ
if Δ is too small, it will not encompass the entire locality
if Δ is too large, it may overlap several localities
if Δ is &infinity;, the working set is the set of all pages touched during process execution
WSS_i is working set size for process p_i
D = Σ WSS_i, where D is the total Demand from frames
if D > m, then thrashing will occur, because some processes will not have enough frames

the OS monitors the working set of each process
and allocates to that working set enough frames to provide it with its working-set size
if there are enough extra frames, a new process can be initiated
if the sum of the working set sizes increases, exceeding the total number of available frames, the OS selects a process to suspend
the working set strategy prevents thrashing while keeping the degree of multiprogramming as high as possible and optimizes CPU utilization.

working set model is successful but keeping track of the working set can become complex
using page-fault frequency (PFF) is a more direct approach to prevent thrashing
if PFF is too high, we know the process needs more frames
if PFF is too low, then we know the process has too many frames
Fig. 9.21