Operating Systems course 20062007 Virtual memory

1
Operating Systems course 2006/2007 Virtual
memory

• Igor Radovanovic

2
Quiz 1 Memory manager

• Memory placement
• Single or multiple (multiprogramming)?
• If multiple processes in memory
• Divide memory in equal sizes or different ones?
• Why do we need memory division?
• After memory is divided
• Can we change sizes of partitions or not
• Dynamic adaptation to process size
• Do processes run in specific partitions or any?
• Should the process be in contiguous blocks or
  not?
• Is the running process entirely in main memory or
  not?
• Which process (or its part) to replace once
  memory is full?

3
Quiz 2 Memory management

• Why do we need primary memory?
• Why do not we use secondary memory only?
• Do you think that the next breakthrough in
  computer systems design would be to make
  processors that can directly fetch instructions
  and data from the secondary memory?
• If YES, .think again
• Since the main memory is becoming huge and
  low-cost, will there be need for memory
  management in the future?
• Cache

4
(OS) MM requirements

• Multi-programming
• Non-contiguous storage
• (parts of) programs stored in main memory ONLY
  when used by CPU
• No memory waste
• No waiting for MM by CPU
• No added complexity
• No additional hardware required
• Unlimited memory size

5
Reminder Memory manager strategies

• Fetch strategies
• Demand fetch strategy
• Anticipatory fetch strategy
• Static placement strategies
• FIFO
• Best-fit
• First-fit
• Worst-fit
• Replacement strategies
• LRU, LFU, other

6
Memory manager implementation

• How is Memory manager implemented?
• Software special-purpose hardware

7
Improved Memory management

• Q
• What needs to be done such that
• programs do not have to be stored contiguously,
  and
• not the entire program, but only its parts need
  to be stored in the memory?
• A
• Split memory into segments and try to fit parts
  of the program into those segments
• If segments are of
• different size segmentation
• the same size paging
• Physical memory blocks frames
• Logical memory blocks pages

8
Memory management-recent systems-

• Paged Memory Allocation
• Demand Paging Memory Allocation
• Segmented Memory Allocation
• Segmented/Demand Paged Allocation

9
Paging memory allocation

• Processes divided into pages of equal size
• Works well if those pages are equal to the memory
  block size (page frame) and the disks section
  size (sectors)
• Advantages of non-contiguous memory allocation
• An empty page frame may be used by any process
• Compaction scheme is not required
• No external and almost no internal fragmentation
• Problems
• Mechanism needed to keep track of page locations
• Consequence Enlarging the size and complexity of
  the OS SW

10
Non-contiguous allocation-an example-
Proc1 size350 lines
Internal fragmentation

• The number of free pages left6
• Any process of more than 600 lines has to wait
  until the Proc1 ends its execution
• Any process of more than 1000 lines cannot fit
  into memory at all!
• Disadvantage
• Entire process must be stored in memory during
  its execution

11
Memory manager using paging requires 3 tables
to track a Procs pages

• 1. Process table (PT) - 2 entries for each active
  process
• Size of a process memory location of its page
  map table
• Dynamic grows/shrinks as processes are
  loaded/completed
• 1 table per job
• 2. Page map table (PMT) - 1 entry per page
• Page number corresponding page frame memory
  address
• Page numbers are sequential (Page 0, Page 1 )
• 1 table per process
• Memory map table (MMT)
• Location free/busy status
• 1 table for all the page frames

12
Process table
13
Page Map Table

• Contains page number and its corresponding page
  frame memory address
• ? - referenced page is not in main memory
• Reading PMT doubles the number of memory access
• Ideally, PMT should not be accessed

14
PMT implementationAssociative memory (AM)

• Used to speed up process execution
• Associative Memory Page Table
• No empty entries (saves memory space)
• Takes advantage of temporal locality and spatial
  locality
• Uses a key-field to address data
• Expensive to implement

15
PMT implementationtranslation-lookaside buffer
(TLB)

• Special registers where page - page frame mapping
  is done
• TLB resides in cache
• The full-page table is in main memory
• When a page is first translated to a page frame,
  the map is read from main memory to TLB
• On subsequent references to the page
• Search for a certain page done in parallel
  through PMTs on the one hand and TLB on the other
• No waste of time if page is not in TLB
• Disadvantage
• High cost of complex hardware required for
  parallel searches

16
TLB requirements

• TLB is the most performance-critical component of
  a modern microprocessor
• Designed in hardware
• Must be
• quickly accessible
• to shorten a clock cycle
• large enough
• to provide avoiding of PMT usage
• Contradictory requirements

17
TLB implementation

• Separate TLBs for instruction fetches and data
  loads
• 2-level TLB (similar to cache)
• Example AMD Opteron
• 40-entry L1 instruction data 512 entry L2
  instruction data
• Take scheduler time slices into account
• What if TLBs are of the same size?
• When should TLB be updated?
• Does this require support of a software in OS?

18
Displacement

• Displacement (offset) of a line -- how far away
  a line is from the beginning of its page
• Used to locate that line within its page frame
• Relative factor
• For example, lines 0, 100, 200, and 300 are first
  lines for pages 0, 1, 2, and 3 respectively, so
  each has displacement of zero
• Address of a given program line
• Division made in hardware
• Memory Management Unit (MMU)

19
Address resolution

• Translate a processs space address into a
  physical address

Page frame no.
Process1 (page size 512)
Main memory
PMT for Process1
20
Memory managementRecent systems

• Paged Memory Allocation
• Demand Paging Memory Allocation
• Segmented Memory Allocation
• Segmented/Demand Paged Allocation

21
Demand paging

• The first allocation scheme that removed the
  restriction of having the entire process stored
  in memory
• Bring a page into memory only when it is needed,
  so less I/O memory needed
• Faster response
• Made virtual memory widely available
• Gives appearance of an almost infinite physical
  memory
• Takes advantage that programs are written
  sequentially so not all pages are necessary at
  once. For example
• User-written error handling modules
• Mutually exclusive modules
• Certain program options are either mutually
  exclusive or not always accessible
• Many tables assigned a fixed amount of address
  space even though only a fraction of a table is
  actually used

22
Demand paging (cntd)

• How and when pages are passed (or swapped)
  depends on predefined policies that determine
  when to make room for needed pages and how to do
  so
• Policy that selects page to be removed is crucial
  to system efficiency
• Selection of algorithm is critical
• The one with the lowest page-fault rate is the
  most attractive
• First-in first-out (FIFO) policy
• best page to remove is one that has been in
  memory the longest
• Least-recently-used (LRU) policy
• chooses pages least recently accessed to be
  swapped out
• Most recently used (MRU) policy
• Least frequently used (LFU) policy

23
Page Map Table
Extra fields with respect to PMT in static paging

• Status bit indicates if page is currently in
  memory or not
• Referenced bit indicates if page has been
  referenced recently
• Used by LRU to determine which pages should be
  swapped out
• Modified bit indicates if page contents have been
  altered
• Used to determine if page must be rewritten to
  secondary storage when it is swapped out
• Note
• Swapping
• copying the page from secondary to main memory

24
Instruction Processing Algorithm

• Hardware components (MMU) do
• generate the (logical) address of the required
  page
• find the page number
• determine if the page is already in memory

• Start processing instruction
• Generate data address
• Compute page number
• if (page is in memory) then
• get data and finish instruction
• advance to next instruction
• return to 1
• else
• generate page interrupt
• call page fault handler
• fi

Page in the secondary storage OS takes over
25
Page fault handler

• Determines whether there are empty page frames in
  memory
• If not, it must decide which page to swap out
• This depends on predefined policy for page
  removal
• When page is swapped out, 3 tables to be updated
• PMT tables of two tasks (1 in,1 out)
• Memory Map Table
• Problem with swapping thrashing

26
Thrashing

• When a process is spending more time paging than
  executing
• When it happens?
• When the number of frames for the low priority
  process falls below the min number required by
  computer architecture
• Pages in active use are replaced by other pages
  in active use
• Happens with increased degree of multiprogramming
  
• Solution
• Local replacement algorithms (not completely)
• Provide the process with as many frames as it
  needs
• Use locality model (working set)
• Note this model also used in caching

27
Thrashing-an example-

• for (j1 jlt100 j)
• kjj
• maj
• printf(\nd d d, j, k, m)
• printf(\n)

Page 0
Swapping between these two pages
Page 1
28
Dynamic allocation paging-Working-set model-

• A set of active pages large enough to avoid
  trashing
• Use a parameter ? to determine a working-set
  window
• Page reference window
• .2 6 1 5 7 7 7 7 5 1 6 2 3 4 1 2 3 4 4 4 3 4 3 4
  4 4 1 3 2 3 4 4 4.
• ?10
  ?10
• WS (t1)1,2,5,6,7
  WS (t2)3,4
• D-the total demand for frames
• WSSi- the working set size
• For thrashing will occur m-the
  total number of available frames

29
A reminder Page fault handler

• Determines whether there are empty page frames in
  memory
• If not, it must decide which page to swap out
• This depends on predefined policy for page
  removal
• Good policy more memory hits - improved
  efficiency
• When page is swapped out 3 tables to be updated
• PMT tables of two tasks (1 in 1 out)
• Memory Map Table
• Problem with swapping thrashing

30
FIFO policy

• How each page requested is swapped into 2
  available page frames using FIFO. When program is
  ready to be processed all 4 pages are on
  secondary storage. Throughout program, 11 page
  requests are issued. When program calls a page
  that isnt already in memory, a page interrupt is
  issued (shown by ). 9
  page interrupts result.

31
FIFO policy-analogy-

• Dresser drawer filled with sweaters
• Situation
• The drawer is full and you have just bought a new
  sweater
• Must put one sweater away to make space for the
  new one
• Q
• Which sweater to remove, oldest or the least
  recently used one?
• Case 1 Remove the oldest one
• What if it turn out that that is your most
  treasured possession?
• Case 2 Remove the least recently used
• What if once-a-year occasion demands its
  appearance soon?
• Case 3 Buy a new wardrobe?!

32
LRU policy

• Throughout the program 11 page requests are
  issued, but they cause only 8 page interrupts
• Efficiency slightly better than FIFO
• The most widely used static replacement algorithm

33
LRU implementation in hardware

• Problem
• Keeping track of of the backward distance of
  every loaded page for every process is difficult
• Page table requires entry for the virtual time of
  the last reference
• Costly
• since another page table memory write must be
  introduce
• Virtual time must be maintained
• Solution
• Approximate LRU behavior
• Check reference bits
• Periodically set to 0
• Introduce shift register with the most
  significant bit like the reference bit and
  periodically shift the content to the right

34
Pros Cons of Demand Paging

• First scheme in which a process was no longer
  constrained by the size of physical memory
  (virtual memory)
• Uses memory more efficiently than previous
  schemes because sections of a process used seldom
  or not at all arent loaded into memory unless
  specifically requested
• Increased overhead caused by tables and page
  interrupts

35
Evaluation of Paging
36
Memory managementRecent systems

• Paged Memory Allocation
• Demand Paging Memory Allocation
• Segmented Memory Allocation
• Segmented/Demand Paged Allocation

37
REMINDER Thrashing-an example-

• for (j1 jlt100 j)
• kjj
• maj
• printf(\nd d d, j, k, m)
• printf(\n)

Page 0
Swapping between these two pages
Page 1
38
Segmented memory allocation

• Based on common practice by programmers of
  structuring their programs in modules (logical
  groupings of code)
• A segment is a logical unit such as main
  program, subroutine, procedure, function, local
  variables, global variables, common block, stack,
  symbol table, or array
• Main memory is not divided into page frames
  because size of each segment is different
• Memory is allocated dynamically
• Address specifying a segment name and an offset
• This in contrast to paging where user specifies
  only a single address

39
Segment Map Table (SMT)

• Memory Manager needs to track segments in memory
• Process Table (PT) lists every process (one for
  the whole system)
• Segment Map Table (SMT) lists details about each
  segment (one for each process)
• Memory Map Table (MMT) monitors allocation of
  main memory (one for the whole system)
• Each segment is numbered and a Segment Map Table
  (SMT) is generated for each process
• Contains segment numbers, their lengths, access
  rights, status, and (when each is loaded into
  memory) its location in memory

40
SMT-an example-

• Two dimensional addressing scheme
• Address (segment number, displacement)

0
0
Main Program
Operating System
3000
Empty
4000
349
Main Program
0
Yin memory Nnot in memory EExecute only
Subroutine A
Other Programs
7000
199
Subroutine A
0
Subroutine B
Other Programs
99
41
Segments vs. Pages

• Pages
• Physical units invisible to the users program
• Fixed size
• Disadvantage
• MM operating without any a priory
  knowledge regarding relationship among pages

• Segments
• Logical units visible to the users program
• Variable size
• Advantages
• Less swapping
• Disadvantages
• External fragmentation
• Memory compaction
• Secondary storage handling

42
Memory managementRecent systems

• Paged Memory Allocation
• Demand Paging Memory Allocation
• Segmented Memory Allocation
• Segmented/Demand Paged Allocation

43
Segmented/Demand Paged Memory Allocation

• Evolved from combination of segmentation and
  demand paging
• Logical benefits of segmentation
• Physical benefits of paging
• Subdivides each segment into pages of equal size,
  smaller than most segments, and more easily
  manipulated than whole segments
• Eliminates many problems of segmentation because
  it uses fixed length pages
• Disadvantage
• Overhead required for extra tables
• Time required to reference both segment table and
  page table

44
Segmented/Demand Paged Memory Allocation -an
example-
Main memory
Page Map Tables Process0/Segment0
0123456789
Segment Map Tables Process0
Active Process Table
Process0/Segment1
Process0/Segment2
Process1
Process1/Segmen0

• Interaction among Process Table, Segment Map
  Table, Page Map Table and Main Memory

45
Managing insufficient memory

• Consolidating several dispersed single holes into
  a larger hole
• Memory compaction
• Swapping
• Overlay

46
Swapping

• Evicting one of the processes (with blocked
  thread) resident in the main memory to the
  secondary storage
• Process manager - Memory manager cooperation
• Make use of the relocation hardware
• Issue where to keep swapped processes on the
  disk?
• Involve Device manager
• In time-sharing systems combined with scheduling
• Scheduling performed at the end of each time
  quantum
• Processes swapped during the execution of other
  processes

47
Swapping (cntd)

• Overhead RS/D the time to reacquire primary
  memory
• R the number of disk blocks
• S the number of units of primary memory
• D - the unit size of the disk block
• Resource waist SxT
• T the amount of time during which the process
  is blocked
• Not always easy to estimate
• Note
• If TltR then process begins requesting memory
  before it is completely swapped out
• Swapping strategy
• Calculate SxT (space-time product),
• T the amount of time the process has held S
  units of memory
• replace the one with the largest product

48
Swapping vs. Memory Compaction

• Advantages
• More efficient
• Guarantees creation of a hole large enough to
  accommodate a request
• Evict as many processes as necessary
• Affects only a small number of processes (not
  always)
• Fewer memory access
• Can be used with both fixed and variable
  partitions
• Disadvantage
• Requires access to secondary memory (time
  consuming)
• Overlapped with execution of other processes
  though

49
Virtual Memory (VM)

• Gives appearance that programs are being
  completely loaded in the main memory during their
  entire processing time
• Shared programs and subroutines are loaded on
  demand, reducing storage requirements of main
  memory
• Spatial locality
• The ideal operation of VM manager
• Knowledge of the exact set of addresses that need
  to be loaded into the main memory before they are
  referenced
• Keeps exactly those addresses in the main memory
  that are currently referenced
• IMPLEMETATION
• VM is implemented through demand paging and
  segmentation schemes

50
VM (cntd)
51
Advantages of VM

• Works well in a multiprogramming environment
  because most programs spend a lot of time waiting
• Processs size is no longer restricted to the
  size of main memory (or the free space within
  main memory)
• Memory is used more efficiently
• Allows an unlimited amount of multiprogramming
• Eliminates external fragmentation when used with
  paging and eliminates internal fragmentation when
  used with segmentation
• Allows a program to be loaded multiple times
  occupying a different memory location each time
• Allows sharing of code and data
• Facilitates dynamic linking of program segments

52
Disadvantages of VM

• Increased processor hardware costs
• Increased overhead for handling paging interrupts
• Increased software complexity to prevent thrashing

53
Cache memory
bus

• Based on the principle of locality
• A high-speed memory that speeds up the CPUs
  access to information
• Increases the performance
• No need to contend for the bus

54
Cache memory-improved efficiency-

• Cache hit ratio h
• Average main memory access time

55
Address translation in Pentium

• Virtual address is 48 bits long (2 bits for
  protection)
• Segment table has two parts of 4 K entries each
  (Local/Global Descriptor Table)

56
The buddy system (in Linux)
Trade-off between fixed and variable partitions
A list with 5 entries (in Linux 10)