Background

• Virtual memory – program uses virtual memory which can be partially

loaded into physical memory

• Benefits:
– Only part of the program needs to be in memory for execution
• more concurrent programs
– Logical address space can therefore be much larger than physical
address space
• execute programs larger than RAM size
– Easy sharing of address spaces by several processes
• Library or a memory segment can be shared
– Allows for more efficient process creation

1
 Virtual Memory That is Larger Than
Physical Memory
Page 0
0
Page 1 1
2 Page
3 Page22 Page 0 Page 1
Page 2
4 unavail
unavail
Page 3 Page 2 Page 3
Page
Page00 move
Page 4 pages
… unavail
unavail Page 4

… Page
Page33
n-2 Page
n-1 Page11 page n-2 Page n-1
page table Physical memory
page n-2
page n-1 all pages of program sitting on physical Disk
Virtual memory
2
A typical virtual-address space layout of a
process
function parameters;
local variables;
return addresses

unused address space

will be used whenever

malloc() allocates needed
space from here
(dynamic memory
allocation)

global data (variables)

3
 Shared Library Using Virtual Memory

Virtual memory of process A Virtual memory of process B

only one copy of

a page
needs to
be in memory

4
 Implementing Virtual Memory

• Virtual memory can be implemented via:

– Demand paging
• Bring pages into memory when they are used, i.e. allocate memory for
pages when they are used

– Demand segmentation
• Bring segments into memory when they are used, i.e. allocate memory
for segments when they are used.

5
 Demand Paging

• Bring a page into memory only when it is needed

– Less I/O needed
– Less memory needed
– Faster response
– More users

• Page is needed  reference to it

– invalid reference (page is not in used portion of address space)  abort
– not-in-memory  bring to memory

• Pager never brings a page into memory unless page will be needed

6
 Valid-Invalid Bit

• With each page table entry a valid–invalid bit is associated

(v  in-memory, i  not-in-memory)
• Initially valid–invalid bit is set to i on all entries
• Example of a page table snapshot: Frame # valid-invalid bit
v
v
v
v
i
….

i
i
page table
• During address translation, if valid–invalid bit in page table entry
is i  page fault
7
Page Table When Some Pages Are Not in Main
Memory

8
 Page Fault

• When CPU makes a memory reference (i.e. page reference), HW consults the
page table. If entry is invalid, then exception occurs and kernel gets executed.
Kernel handling such as case:
1. Kernel looks at another table to decide:
– Invalid reference (page is in unused portion of address space)  Abort
– Just not in memory (page is in used portion, but not in RAM)  Page Fault
2. Get empty frame
(we may need to remove a page; if removed page is modified, we need disk
I/O to swap it out)
3. Swap page into frame
(we need disk I/O)
4. Reset tables (install mapping into page table)
5. Set validation bit = v
6. Restart the instruction that caused the page fault

9
 Page Fault (Cont.)

• If page fault occurs when trying to fetch an instruction, fetch the

instruction again after bringing the page in.
• If page fault occurs while we are executing an instruction: Restart the
instruction after bringing the page in.

• For most instructions, restarting the instruction is no problem.

– But for some, we need to be careful.

10
Steps in Handling a Page Fault

swap
space

11
 Performance of Demand Paging

• Page Fault Rate (p): 0  p  1.0

– if p = 0 no page faults
– if p = 1, every reference is a fault

• Effective Access Time to Memory (EAT)

EAT = (1 – p) x memory_access_time
+ p x (page fault overhead time
+ time to swap page out (sometimes)
+ time swap page in
+ restart overhead time)
page fault
service time

12
 Demand Paging Example

• Memory access time = 200 nanoseconds

• Average page-fault service time = 8 milliseconds

• EAT = (1 – p) x 200 + p (8 milliseconds)

= (1 – p) x 200 + p x 8,000,000
= 200 + p x 7,999,800

• If one access out of 1,000 causes a page fault (p = 1/1000), then

EAT = 8.2 microseconds.

This is a slowdown by a factor of 40!!

(200 ns / 8.2 microsec ~= 1/40)

13
 Process Creation

• Virtual memory allows other benefits during process creation:

- Copy-on-Write

- Memory-Mapped Files (later)

14
 Copy-on-Write

• Copy-on-Write (COW) allows both parent and child processes to initially share
the same pages in memory

If either process modifies a shared page, only then is the page copied

• COW allows more efficient process creation as only modified pages are copied

15
Before Process 1 Modifies Page C

16
After Process 1 Modifies Page C

17
Page Replacement

18
 What happens if there is no free frame?

• Page replacement – find some page in memory, but not really in use, swap it
out

– Algorithm ? Which page should be remove?

– performance – want an algorithm which will result in minimum number of

page faults

• With page replacement, same page may be brought into memory several
times

• Prevent over-allocation of memory by modifying page-fault service routine to

include page replacement

19
 Page Replacement

• Use modify (dirty) bit to reduce overhead of page transfers – only modified
pages are written to disk while removing/replacing a page.

• Page replacement completes separation between logical memory and physical

memory
– large virtual memory can be provided on a smaller physical memory

20
Need For Page Replacement

While executing “load M”

we will have a page
fault and we need
page replacement.
21
 Basic Page Replacement

Steps performed by OS while replacing a page upon a page fault:

1. Find the location of the desired page on disk

2. Find a free frame:

- If there is a free frame, use it
- If there is no free frame, use a page replacement algorithm to select a
victim frame; if the victim page is modified, write it back to disk.

3. Bring the desired page into the (new) free frame; update the page and frame
tables

4. Restart the process

22
Page Replacement

23
 Page Replacement Algorithms

• Want lowest page-fault rate

• Evaluate algorithm by running it on a particular string of memory references

(reference string) and computing the number of page faults on that string

• In all our examples, the reference string is

1, 2, 3, 4, 1, 2, 5, 1, 2, 3, 4, 5

24
 Driving reference string

• Assume process makes the following memory references (in decimal) in a

system with 100 bytes per page:

• 0100 0432 0101 0612 0102 0103 0104 0101 0611 0102 0103 0104
0101 0610 0102 0103 0104 0609 0102 0105
• Example: Bytes (addresses) 0…99 will be in page 0
• Pages referenced with each memory reference
– 1, 4, 1, 6, 1, 1, 1, 1, 6, 1, 1, 1, 1, 6, 1, 1, 6, 1, 1

• Corresponding page reference string

– 1, 4, 1, 6, 1, 6, 1, 6, 1, 6, 1

25
Graph of Page Faults Versus The Number of
Frames

26
 First-In-First-Out (FIFO) Algorithm
• Reference string: 1, 2, 3, 4, 1, 2, 5, 1, 2, 3, 4, 5
• 3 frames (3 pages can be in memory at a time per process)

1 1 4 5
2 2 1 3 9 page faults
3 3 2 4
• 4 frames

1 1 5 4
2 2 1 5 10 page faults
3 3 2

4 4 3

• Belady’s Anomaly: more frames  more page faults

27
FIFO Page Replacement

28
FIFO Illustrating Belady’s Anomaly

29
 Optimal Algorithm

• Replace page that will not be used for longest period of time
• 4 frames example
1, 2, 3, 4, 1, 2, 5, 1, 2, 3, 4, 5

1 4
6 page
2
faults
3

4 5

• How do you know this?

• Used for measuring how well your algorithm performs

30
Optimal Page Replacement

31
Least Recently Used (LRU) Algorithm

• Reference string: 1, 2, 3, 4, 1, 2, 5, 1, 2, 3, 4, 5

1 1 1 1 5
2 2 2 2 2
3 5 5 4 4
4 4 3 3 3

8 page faults

32
LRU Page Replacement

33
 LRU Algorithm Implementation

• Counter implementation

– Every page entry has a counter field; every time page is referenced
through this entry, copy the clock into the counter field

– When a page needs to be replaced, look at the counters to determine

which one to replace
• The one with the smallest counter value will be replaced

34
 LRU Algorithm Implementation

• Stack implementation – keep a stack of page numbers in a double link form:

– Page referenced:
• move it to the top
• requires 6 pointers to be changed (with every memory reference;
costly)
– No search for replacement (replacement fast)

35
Use of a Stack to Record The Most Recent Page
References

36
 LRU Approximation Algorithms

• Reference bit
• Additional Reference bits
• Second Chance 1
• Second Chance 1 (clock)
• Enhanced Second Chance

37
 Reference Bit

• Use of reference bit

– With each page associate a bit, initially = 0 (not referenced/used)
– When page is referenced, bit set to 1
– Replace the one which is 0 (if one exists)
• We do not know the order, however (several pages may have 0 value)

– Reference bits are cleared periodically

• (with every timer interrupt, for example);

38
 Additional Reference Bits

• Besides the reference bit (R bit) for each page, we can keep an
AdditionalReferenceBits (say ARB) field associated with each page. For
example, an 8-bit field that can store 8 reference bits.

• At each timer interrupt (or periodically), the reference bit of a page is shifted
from right to the AdditionalReferenceBits field of the page. All other bits of
AdditionalReferenceBits field is shifted to the right as well.
– The value in the AdditionalReferenceBits field will indicate when the page
is accessed (referenced), approximately.

• When a page is to be replaced, select the page with least

AdditionalReferenceBits field value.

39
 Additional Reference Bits
Example
• At tick 1: R: 0, ARB: 0000000
• R is set (R:1)
• At tick 2: R:0, ARB: 1000000
• R is not set
• At tick 3: R:0, ARB: 0100000
• R is set (R:1)
• At tick 4: R:0, ARB: 1010000
• ….

40
 Second-Chance Algorithm 1

– FIFO that is checking if page is referenced or not; Need R bit

• If page to be replaced, look to the FIFO list; remove the page close to
head of the list and that has reference bit 0.
– If the head has R bit 1, move it to the back of the list (i.e. set the load time
to the current time) after clearing the R bit.
• Then try to find another page that has 0 as R bit.
– May require to change all 1’s to 0’s and then come back to the beginning
of the queue.
– Add a newly loaded page to the tail with R = 1.

R=1 R=1 R=0 R=0 R=1 R=1

Head Tail
(oldest) (Youngest)

41
 Second-Chance Algorithm 1

• Example: Head

1 1 0 0 1
Before page removal C A B E D

Access page H

0 1 0 0 1
After page removal E D C A H

42
 Second-Chance Algorithm 2
(Clock Algorithm)

Second chance can

be implemented using
a circular list of pages;
Then it is also called
Clock algorithm
Next victim pointer

43
 Enhanced Second-Change Algorithm

• Consider also the reference bits and the modified bits of pages
– Reference (R) bit: page is referenced in the last interval
– Modified (M) bit: page is modified after being loaded into memory

• Four possible cases (R,M):

– 0,0: neither recently used nor modified
– 0,1: not recently used but modified
– 1,0: recently used but clean
– 1,1: recently used and modified
• We replace the first page encountered in the lowest non-empty class.
• Rest is the same with second-chance algorithm

• We may need to scan the list several times until we find the page to replace

44
 Counting Algorithms

• Keep a counter of the number of references that have been made to each
page

• LFU Algorithm: replaces page with smallest count

• MFU Algorithm: based on the argument that the page with the smallest count
was probably just brought in and has yet to be used

45
 Allocation of Frames

• Each process needs minimum number of pages

• Example: IBM 370 – 6 pages to handle SS MOVE instruction:
– instruction is 6 bytes, might span 2 pages
– 2 pages to handle from
– 2 pages to handle to

• Various allocation approaches

– fixed allocation (this is a kind of local allocation)
• Equal allocation
• Proportional allocation (proportional to the size)
– priority allocation (this is a kind of global allocation)
– global allocation
– local allocation

46
 Fixed Allocation
• Equal allocation – For example, if there are 100 frames and 5
processes, give each process 20 frames.
• Proportional allocation – Allocate according to the size of process

si  size of process pi
S   si
m  total number of frames
m  64
si
ai  allocation for pi   m si  10
S
s2  127
Example: 10
a1   64  5
137
127
a2   64  59
137

47
 Priority Allocation

• Use a proportional allocation scheme using priorities rather than size

• If process Pi generates a page fault,

– select for replacement one of its frames
– select for replacement a frame from a process with lower priority number

48
 Global versus Local Allocation

• When a page fault occurs for a process and we need page replacement, there
are two general approaches:

– Global replacement – select a victim frame from the set of all frames;
• one process can take a frame from another

– Local replacement – select a victim frame only from the frames allocated
to the process.
• A process uses always its allocated frames

49
 Thrashing

• If a process does not have “enough” pages, the page-fault rate is very high.
This leads to:
– low CPU utilization
– operating system thinks that it needs to increase the degree of
multiprogramming
– another process added to the system

• Thrashing  a process is busy swapping pages in and out

50
Thrashing (Cont.)

51
 Demand Paging and Thrashing

• Why does demand paging work?

Locality model (locality of reference)
– Process migrates from one locality to another
– Localities may overlap

• Why does thrashing occur?

 size of locality > total memory size

52
Locality In A Memory-Reference Pattern

53
 Working-Set Model

• A method for deciding

a) how many frames to allocate to a process, and also
b) for selecting which page to replace.

• Maintain a Working Set (WS) for each process.

– Look to the past  page references
   working-set window  a fixed number of page references

• WSSi (working set size of Process Pi) = total number of distinct pages
referenced in the most recent 
– WSS varies in time
– Value of  is important
• if  too small will not encompass entire locality
• if  too large will encompass several localities
• if  =   will encompass entire program

54
 Working-Set Model

• D =  WSSi  total demand for frames

• if D > m  Thrashing (m: #frames in memory)

• A possible policy: if D > m, then suspend one of the processes.

55
Working-Set Model

56
 Keeping Track of Working-Set
a method

additional
ref_bits Physical Memory
R_bit (ARB) page x frame 0
x 0 x 0 0
y 0 y 0 0 Page y frame 1
z 0 z 0 0
w 0 w 0 0 Page z frame 2

Page w frame 3
page table

ARB is 2 bits here, but could be more (like 8 bits)

57
 Keeping Track of Working-Set
a method

• Approximate with interval timer + a reference bit

• Example:  = 10,000 (time units)

Timer interrupts after every 5000 time units
Keep 2 bits for each page

Whenever timer interrupts, for a page, shift the R bit from right into
ARB and clear R bit.

If ARB has at least one 1  page in working set

you can increases granularity by increasing the size of ARB and decreasing
the timer interrupt interval

58
Working Sets and Page Fault Rates

transition from one working set to another

59
 Page-Fault Frequency (PFF) Scheme

• Establish “acceptable” page-fault rate

– If actual rate too low, process loses frame
– If actual rate too high, process gains frame

60
 Memory-Mapped Files

• Memory-mapped file I/O allows file I/O to be treated as routine memory access
by mapping a disk block to a page in memory

• A file is initially read using demand paging. A page-sized portion of the file is
read from the file system into a physical page. Subsequent reads/writes
to/from the file are treated as ordinary memory accesses.

• Simplifies file access by treating file I/O through memory rather than read()
write() system calls

• Also allows several processes to map the same file allowing the pages in
memory to be shared

61
Memory Mapped Files

62
Memory-Mapped Shared Memory in
Windows

63
 Allocating Kernel Memory

• Treated differently from user memory. Why?

• Often allocated from a free-memory pool

– Kernel requests memory for structures (objects) of varying sizes
• Object types: process descriptors, semaphores, file objects, …
• Allocation of object type size requested many times.
– Those structures have sizes much less than the page size
– Some kernel memory needs to be contiguous

• This is dynamic memory allocation problem.

• But using first-fit like strategies (heap management strategies) cause external
fragmentation

64
 Allocating Kernel Memory

• We will see two methods

– Buddy System Allocator

– Slab Allocator

65
 Buddy System Allocator

• Allocates memory from fixed-size segment consisting of physically-contiguous

pages

• Memory allocated using power-of-2 allocator

– Satisfies requests in units sized as power of 2
– Request rounded up to next highest power of 2

– When smaller allocation needed than is available, current chunk split into
two buddies of next-lower power of 2
• Continue until appropriate sized chunk available

66
Buddy System Allocator

67
 Example

• Object A needs memory 45 KB in size

• Object B needs memory 70 KB in size
• Object C needs memory 50 KB in size
• Object D needs memory 90 KB in size
•
• Object C removed
• Object A removed
• Object B removed
• Object D removed

68
 Example
512 KB of Memory (physically contiguous area)

A C B D

Alloc A 45 KB
512
Alloc B 70 KB
Alloc C 50 KB
256 256 Alloc D 90 KB
Free C
Free A
128 128(B)
128 128
128(D) 128 Free B
Free D
64(A)
64 64(C)
64

69
 Slab Allocator

• Alternate strategy

• Within kernel, a considerable amount of memory is allocated for a finite set of

objects such as process descriptors, file descriptors and other common
structures

• Idea: a contiguous phy memory (slab) a contiguous phy memory (slab)

(a set of page frames) (a set of page frames)

Obj Obj Obj Obj Obj Obj

Obj Obj
X X X X Y Y Y

Obj X: object of type X

Obj Y: object of type Y

70
 Slab Allocator

• Slab is one or more physically contiguous pages

• Cache consists of one or more slabs
• Single cache for each unique kernel data structure
– Each cache filled with objects – instantiations of the data structure

• When cache created, filled with slots (objects) marked as free

• When structures stored, objects marked as used

• If slab is full of used objects, next object allocated from empty slab
– If no empty slabs, new slab allocated

• Benefits include
– no fragmentation,
– fast memory request satisfaction

71
 Slabs and Caches

cache cache
structure structure

slab slab
structure structure

a set of a set of a set of a set of a set of

contiguous contiguous contiguous contiguous contiguous
pages pages pages pages pages
(a slab) (a slab) (a slab) (a slab) (a slab)

set of slabs containing same type of a set of slabs

objects (a cache) (another cache)
(can store objects of type/size X) (can store objects of type/size Y)
72
Slab Allocation

73
 Prepaging

• Prepaging
– To reduce the large number of page faults that occurs at process startup
– Prepage all or some of the pages a process will need, before they are
referenced
– But if prepaged pages are unused, I/O and memory was wasted

– Assume s pages are prepaged and α of the pages is used

• Is cost of s * α save pages faults > or < than the cost of prepaging
s * (1- α) unnecessary pages?
• α near zero  prepaging loses

74
 Other Issues – Page Size

• Page size selection must take into consideration:

– Fragmentation
• Small page size reduces fragmentation
– table size
• Large page size reduces page table size
– I/O overhead
• Large page size reduce I/O overhead (seek time, rotation time)
– Locality
• Locality is improved with smaller page size.

75
 Other Issues – TLB Reach

• TLB Reach - The amount of memory accessible from the TLB

• TLB Reach = (TLB Size) x (Page Size)

• Ideally, the working set of each process is stored in the TLB

– Otherwise there is a high degree of page faults

• To increase TLB reach:

– Increase the Page Size
• This may lead to an increase in fragmentation as not all applications
require a large page size

– Provide Multiple Page Sizes

• This allows applications that require larger page sizes the opportunity
to use them without an increase in fragmentation

76
 Other Issues – Program Structure

• Program structure page 0 int int int int

– int[128,128] data; Page 1 int int int int
– Each row is stored in one page
assuming pagesize=512 bytes Page 127 int int int int
– Program 1
for (j = 0; j <128; j++)
for (i = 0; i < 128; i++)
data[i,j] = 0;

128 x 128 = 16,384 page faults

– Program 2
for (i = 0; i < 128; i++)
for (j = 0; j < 128; j++)
data[i,j] = 0;

128 page faults

77
 Other Issues – I/O interlock

• I/O Interlock – Pages must sometimes

be locked into memory

Process A
• Consider I/O - Pages that are used for
copying a file from a device must be pages
locked from being selected for eviction
by a page replacement algorithm

Process B
pages

Process A starts I/O and then blocks.

Process B runs and needs a frame.
We should not remove A’s page
78
Additional Study Material

79
 Operating System Examples

• Windows XP

• Solaris

80
 Windows XP

• Uses demand paging with clustering. Clustering brings in pages surrounding

the faulting page
• Processes are assigned working set minimum and working set maximum
• Working set minimum is the minimum number of pages the process is
guaranteed to have in memory
• A process may be assigned as many pages up to its working set maximum
• When the amount of free memory in the system falls below a threshold,
automatic working set trimming is performed to restore the amount of free
memory
• Working set trimming removes pages from processes that have pages in
excess of their working set minimum

81
 Solaris

• Maintains a list of free pages to assign faulting processes

• Lotsfree – threshold parameter (amount of free memory) to begin paging
• Desfree – threshold parameter to increasing paging
• Minfree – threshold parameter to being swapping
• Paging is performed by pageout process
• Pageout scans pages using modified clock algorithm
• Scanrate is the rate at which pages are scanned. This ranges from slowscan
to fastscan
• Pageout is called more frequently depending upon the amount of free memory
available

82
Solaris 2 Page Scanner

83
Slab Allocation in Linux Kernel

84
 Cache structure

• A set of slabs that contain one type of object is considered as a cache.

• Cache structure is a structure that keeps information about the cache and
includes pointers to the slabs.

struct kmem_cache_s {
struct list_head slabs_full; /* points to the full slabs */
struct list_head slabs_partial; /* points to the partial slabs */
struct list_head slabs_free; /* points to the free slabs */
unsigned int objsize; /* size of objects stored in this cache */
unsigned int flags;
unsigned int num;
spinlock_t spinlock;
…
…
}

85
 Slab structure

• A slab stucture is a data structure that points to a contiguous set of page

frames (a slab) that can store some number of objects of same size.
• A slab can be considered as a set of slots (slot size = object size). Each slot
in a slab can hold one object.

• Which slots are free are maintained in the slab structure

typedef struct slab_s {

struct list_head list;
unsigned long colouroff;
void *s_mem; /* start address of first object */
unsigned int inuse; /* number of active objects */
kmem_bufctl_t free; /* info about free objects */
} slab_t;

86
 Layout of Slab Allocator

prev cache next cache

cache

slabs_full slabs_partial slabs_free

slabs slabs slabs

pages pages pages

an object

87
 Slab Allocator in Linux

• cat /proc/slabinfo will give info about the current slabs and objects
cache names: one cache for each different object type
# name <active_objs> <num_objs> <objsize> <objperslab> <pagesperslab> : tunables <limit> <batchcount> <
sharedfactor> : slabdata <active_slabs> <num_slabs> <sharedavail>
ip_fib_alias 15 113 32 113 1 : tunables 120 60 8 : slabdata 1 1 0
ip_fib_hash 15 113 32 113 1 : tunables 120 60 8 : slabdata 1 1 0
dm_tio 0 0 16 203 1 : tunables 120 60 8 : slabdata 0 0 0
dm_io 0 0 20 169 1 : tunables 120 60 8 : slabdata 0 0 0
uhci_urb_priv 4 127 28 127 1 : tunables 120 60 8 : slabdata 1 1 0
jbd_4k 0 0 4096 1 1 : tunables 24 12 8 : slabdata 0 0 0
ext3_inode_cache 128604 128696 504 8 1 : tunables 54 27 8 : slabdata 16087 16087 0
ext3_xattr 24084 29562 48 78 1 : tunables 120 60 8 : slabdata 379 379 0
journal_handle 16 169 20 169 1 : tunables 120 60 8 : slabdata 1 1 0
journal_head 75 144 52 72 1 : tunables 120 60 8 : slabdata 2 2 0
revoke_table 2 254 12 254 1 : tunables 120 60 8 : slabdata 1 1 0
revoke_record 0 0 16 203 1 : tunables 120 60 8 : slabdata 0 0 0
scsi_cmd_cache 35 60 320 12 1 : tunables 54 27 8 : slabdata 5 5 0
….
files_cache 104 170 384 10 1 : tunables 54 27 8 : slabdata 17 17 0
signal_cache 134 144 448 9 1 : tunables 54 27 8 : slabdata 16 16 0
sighand_cache 126 126 1344 3 1 : tunables 24 12 8 : slabdata 42 42 0
task_struct 179 195 1392 5 2 : tunables 24 12 8 : slabdata 39 39 0
anon_vma 2428 2540 12 254 1 : tunables 120 60 8 : slabdata 10 10 0
pgd 89 89 4096 1 1 : tunables 24 12 8 : slabdata 89 89 0
pid 170 303 36 101 1 : tunables 120 60 8 : slabdata 3 3 0

active objects size

88
 References

• The slides here are adapted/modified from the textbook and its slides:
Operating System Concepts, Silberschatz et al., 7th & 8th editions, Wiley.
• Operating System Concepts, 7th and 8th editions, Silberschatz et al. Wiley.
• Modern Operating Systems, Andrew S. Tanenbaum, 3rd edition, 2009.

89