Page Replacement ALgos

7/28/2019 Page Replacement ALgos

1/42

P r l m n l ri hm

1

OS 2009-10


2/42

When a page fault occurs2

OS has to choose a page to evict from memory

If the page has been modified, the OS has to schedulea disk write of the page

e page us rea overwr es a page n memory e.g.4Kbytes)

,

Same problem applies to memory caches

OS 2009-10


3/42

Benchmarking3

Tests are done b eneratin a e references (either

from real code or random) Sequences of page numbers (no real address, no

offset)

Example:

7 0 1 2 0 3 0 4 2 3 0 3 2 1 2 0 1 7 0 1

OS 2009-10


4/42

Optimal page replacement4

At the moment of page fault:

Label each page in memory is labeled with the number ofinstructions that will be executed before that page is firstreferenced

Replace the page with the highest number: i.e. postpone asmuch as possible the next page fault

, , The OS cant look into the future to know how long itll take to

reference every page again

OS 2009-10


5/42

Example: optimal5

Sequence

6 page faults

OS 2009-10


6/42

Beladys anomaly6

Try this sequence

With 3 page frames

With FIFO, with the optimal algorithm, (later) with the LRU

OS 2009-10


7/42

Not recently used algorithm7 Use Referenced and Modified bits

R&M are in hardware, potentially changed at eachreference to memory

On clock interrupt the Rbit is cleared

On page fault, to decide which page to evict: Classify:

Class 0 R=0,M=0 Class 1 R=0,M=1

Class 2 R=1,M=0 Class 3 R=1,M=1

Replace a page at random from the lowest class

OS 2009-10


8/42

FIFO replacement8

FIFO, first in first out for a es

Clearly not particularly optimal It mi ht end u removin a a e that is still

referenced since it only looks at the pages age

Rarely used in pure form

1 8 3 7 2

Latest load

OS 2009-10

Next removal


9/42

Example (FIFO)9

Sequence

PF

12 page faults

OS 2009-10


10/42

Second chance algorithm10

Like FIFO but

Before throwing out a page checks the R bit: If 0 remove it

If 1 clear it and move the page to the end of the list (as it werejust been loaded)

= ,FIFO (why?)

1 8 3 7 2

OS 2009-10

Latest load


11/42

Clock page algorithm11

implemented differently: Check startin from the

latest visited page

More efficient:

ba

lists entries all thetime dg

ef

OS 2009-10


12/42

Least recently used (LRU)12

Why: pages recently used tend to be used again

soon (on average)

List of all pages in memory (most recently in thehead, least recently used in the tail)

List operations are expensive (e.g. find a page)

So, difficult

OS 2009-10


13/42

Least recently used (LRU)13

Idea! Get a counter, maybe a 64bit counter

The counter is incremented after each instructionand stored into the page entry at each reference

Store the value of the counter in each entry of thepage table (last access time to the page)

,counter value (this is the LRU page)

Nice & good but expensive: it requires dedicatedhardware

OS 2009-10


14/42

Example LRU14

Sequence

PF

9 page faults

OS 2009-10


15/42

NFU algorithm15

Since LRU is ex ensive

NFU: Not Frequently Used algorithm At each clock interru t add the R bit to a counter for

each page: i.e. count how often a page is referenced

Remove page with lowest counter value Unfortunately, this version tends not to forget

anything

OS 2009-10


16/42

Aging (NFU + forgetting)16

Take NFU but

At each clock interrupt: Ri ht shift the counters (divide b 2)

Add the R bit to the left (MSB)

As for NFU remove pages with lowest counter

Note: this is different from LRU since the time

granularity is a clock tick and not every memoryreference!

OS 2009-10


17/42

NFU+ageing17

OS 2009-10


18/42

Process behavior18

Locality of reference: most of the time the last kreferences are within a finite set of pages < a largeaddress space

the working set(WS) of the process

Knowin the workin set of rocesses we can do versophisticate things (e.g. pre-paging)

Pre-paging: load pages before letting the process to

run so a e page- au ra e s ow, a so, nowthe WS when I swap the process then I can expect itto be the same in the future time when I reload the

OS 2009-10

process in memory


19/42

Working set19

set

Working

k-most-recent memory references

OS 2009-10


20/42


21/42

e.g.: WS based algorithm21

OS 2009-10


22/42

WSClock algorithm22

Use the circular structure (as seen earlier)

At page fault examine the page pointed by the handle

R=1, page in the WS dont remove it (set R to zero)

, , ,and decide depending on age)

If M=1, schedule disk write appropriately to procrastinateas ong as poss e a process sw c If return to starting point, then one page will eventually be clean

(maybe after a context switch)

c e u ng mu t p e s wr te can e e c ent n e c ent systems(with disk scheduling and multiple disks)

No write is schedulable (R=1 always), just choose a clean page

OS 2009-10

o c ean page, use e curren page un er e an e


23/42

WSClock23

R=1

R=0

OS 2009-10


24/42


25/42

25

Design issues

OS 2009-10


26/42

Design issues26

Local vs. lobal allocation olic

When a page fault occurs, whose page should the OS evict? Which process should get more or less pages? Monitor the number of page faults for every process (PFF

page fault frequency)

,less page faults

OS 2009-10


27/42


28/42

Load control28

If the WS of all rocesses > memor , theres

thrashing E.g. the PFF says a process requires more memory

but none require less

Solution: swapping swap a process out of memoryan re-ass gn ts pages to ot ers

OS 2009-10


29/42

Page size29

Pa e size n a es of memor

Average process size s, in pages s/p Each entr in the a e table re uires e tes

On averagep/2 is lost (fragmentation)

used within pages

Wasted memor : 2 + se

Minimizing it yields the optimal page size (undersim lif in assum tions

OS 2009-10


30/42


31/42

Other issues31

Shared a es handle shared a es e. . ro ram

code) Sharing data (e.g. shared memory)

Cleaning policy

Paging algorithms work better if there are a lot of freepages ava a e

Pages need to be swapped out to disk

and evict pages if there are to few)

OS 2009-10


32/42

Page fault handling32

1. Page fault, the HW traps to the kernel. . .

2. Save general purpose microprocessor information (registers, PC, PSW, etc.)3. The OS looks for which page caused the fault (sometimes this information is

already somewhere within the MMU).

protection fault is generated, and the process killed)5. The OS looks for a free page frame, if none is found then the replacement

algorithm is run. =calling process)

7. When the page frame is clean, the OS schedules another transfer to read in therequired page from disk

. ,9. The faulting instruction is backed up, the situation before the fault is restored,

the process resumes execution

OS 2009-10


33/42


34/42

Why?34

Many separate address spaces (segments) (e.g. data, stack,co e, an many ot ers nee e

Each segment is separate (e.g. addresses from 0 to someMAX)

Segments might have different lengths

Segment number + address within segment n ng s s mp e rar es w n eren segmen scan assume addresses starting from 0) e.g. if a part of thelibraries is recompiled the remainder of the code is

una ecte Shared library (DLLs) implementation is simpler (the

sharin is sim ler)

OS 2009-10


35/42

Comparing paging and segmentation35

Need the programmer be aware that

this technique is being used?

No Yes

How many linear address spaces are 1 Many

there?

Can the total address space exceed

the size of physical memory

Yes Yes

Can procedures and data be No Yes

distinguished and separately

protected?

Can tables whose size fluctuate be

accommodated easily?

No Yes

s s ar ng o proce ures e ween

users facilitated?

o es

Why was this technique invented? To get a large linear address space

without having to buy more physical

memory

To allow programs and data to be

broken up into logically independent

address spaces and to aid sharing

OS 2009-10

and protection


36/42

Pure segmentations36

B B B

A A AD

Operating system Operating system Operating system Operating system

OS 2009-10


37/42


38/42


39/42

The segment descriptor39

This is used by the microcode within the Pentiumto work with segments

Limit in pages/bytes

Base 24-31 G D 0 Limit 16-19 P DPL S Type Base 16-23

16/32 bit segment

Privilege level Segment type

protection

Base 0-15 Limit 0-15

Segment present in memory

Page size is 4K System/application

Limit (20 bits)

OS 2009-10

n ex r v ege

Selector


40/42


41/42


42/42

More on the Pentiums42

TLB, to avoid re eated accesses to memor

The whole thing can be used with just a singlesegment to obtain a linear 32bit address space

Set base and limit appropriately

Protection (a few bits)

OS 2009-10

Documents

Page Replacement ALgos