Chapter 7: Memory Management

1
Chapter 7 Memory Management

• CS 472 Operating Systems
• Indiana University Purdue University Fort Wayne

2
Memory management

• In a uniprogramming system, memory is divided
  into one partition for the OS and another
  partition for the user program
• In a multiprogramming system, the user area is
  subdivided so each process has its own partition
• This allows multiple processes to run
  concurrently
• A strategy for doing this is known as memory
  management

3
Memory management

• Memory should be efficiently allocated so that a
  maximum number of processes can coexist
• We want to have at least one process in the ready
  queue when the processor becomes available
• Some requirements
• Ability to relocate a program
• Each partition should be protected
• Processes should be able to establish shared
  memory areas

4
Review of addressing
absolute address

• Relative address
• an offset from the load address
• Absolute address
• address of a physical memory location
• used, for example, with memory-mapped I/O

load address
program
relative address
5
Early systems

• Compiled programs in early systems used only
  absolute addresses
• The load address needed to be specified at
  compile time
• Data areas were allocated within the load module
• If the program was loaded elsewhere, references
  would be off
• Hard to combine separately compiled program units

6
Later systems with relative addressing

• Alinker program is introduced to combine
  separately compiled program units
• The linker needs to modify relative addresses of
  each separate unit to reflect the load address of
  the first unit
• Linker output is an .exe file
• Absolute addresses (say, for memory-mapped I/O)
  are not changed

7
Linker
myProgram.exe
load address
unit1.obj
modified relative addresses
unit2.obj
unit3.obj
8
Protection

• A process should not be able to reference memory
  locations in another process without permission
• Hardware support needed for protection
• Not possible to check dynamically calculated
  addresses at compile time
• All addresses must be checked at run time

9
Hardware support for relative addressing
interrupt
relative address

compare
absolute address
base register
bounds register
load address
load address size
10
Hardware and relative addressing . . .

• Allows a program to be built out of separately
  compiled units
• Bounds register provides protection to other
  users
• Allows the program to be swapped out of memory
  and then swapped back in at a different load
  address
• This is called dynamic relocation

11
Memory management techniques

• Fixed partitioning
• Dynamic partitioning
• Buddy system
• Simple paging
• Simple segmentation
• Virtual memory with paging
• Virtual memory with segmentation

In order of increasing sophistication
12
Fixed partitioning
memory

operating system

• The number and size of partitions is fixed
• The sizes can be the same or different
• A program must use overlays if too big
• Internal fragmentation results if program too
  small
• This is wasted space at the end of a partition
• See next slide

13
Fixed partitioning

operating system

• Placement algorithm for partitions of mixed size
• Feed all new processes from a single queue
• For the next process, allocate the smallest
  available partition that will hold the process

process 2
process 1
process 3
fragmentation
14
Dynamic partitioning
memory

operating system

• No fixed number of partitions
• Partition size process size
• no internal fragmentation
• External fragmentation
• As processes come and go, many scattered small
  holes grow between processes
• See next slide

process 1
process 2
process 3
process 4
External fragmentation
15
Dynamic partitioning
memory

operating system

• Compaction needed
• time consuming
• requires dynamic relocation
• Placement algorithm choices
• best-fit, first-fit, and next-fit
• First-fit is simplest and best
• Best-fit leads to smaller fragments
• These are usually too small to use

process 5
process 6
process 2
process 7
process 4
External fragmentation
16
Buddy system

• A compromise between the rigidity of fixed
  partitioning and the complexity and overhead of
  dynamic partitioning
• Allocates blocks of size 2k
• Tries to allow a variety of block sizes while
  avoiding excess internal fragmentation

17
Buddy system

• Initial memory is one block of 2U bytes
• For a memory request of size s, find the smallest
  i such that 2i-1lt s 2i
• Successively divide blocks in half until size 2i
  is reached
• Allocate one block, save the buddy
• Maintain a list of holes of each size
• When two buddies of size 2i become available,
  merge them into a hole of size 2i1

18
(No Transcript)
19
(No Transcript)
20
Simple paging

• Partition memory into page frames of fixed
  equal size
• Divide each process into pages of the same size
  as page frames
• Any page may be loaded into any page frame
• Simple means all pages of a process must be
  loaded into page frames when the process is
  started

21
Simple paging

• The OS maintains a page table for each process
• Table consists of page table entries (PTEs)
• Entry n in the table is the PTE of page n
• The PTE of page n gives the frame number where
  the page is loaded
• Internal fragmentation exists only on the last
  page of each process
• No external fragmentation

22
Assignment of pages to free page frames
23
Assignment of pages to free page frames
24
Corresponding page tables
25
Address translation

• Must be done automatically by hardware
• Uses the page tables
• Each logical address consists of two parts
• page number
• offset within the page

logical address
page number offset
26
Address translation

• Consider a logical address of nm bits
• leftmost n bits are the page number
• rightmost m bits are the offset
• Use the page number as an index into the process
  page table
• Obtain the bits of the frame number from the PTE
• Append the offset to the frame number and obtain
  the physical address that corresponds to the
  logical address

27
(No Transcript)
28
Simple segmentation

• Similar to simple paging except programmer
  modularizes process into segments s0, s1, s2, . .
  .
• Segments may be of different sizes (smaller than
  a maximum size)
• Each segment is loaded into a dynamic partition
• Each segment has a load address and a length
• Simple means that all segments must be loaded
  into partitions when the process starts
• External fragmentation exists

29
Address translation

• The OS maintains a segment table for each
  process
• Address translation must be done automatically by
  hardware
• Each logical address consists of two parts
• segment number
• offset within the segment
• If the offset exceeds the segment length, the
  process is terminated with a memory violation

30

• Address translation

31
Appendix 7A

• Good review of linking and loading
• See pages 326 - 331