Memory Management
DVGBO1: Operativsystem
Mohammad Rajiullah
kau.se/cs
Based on the slides from book and Umass OS
•
•
•
Memory Mangement
Program needs to be in memory for CPU access
Multiple programs share memory that need to be coordinated
Most solutions require hardware support.
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Background
• Program executable starts out on disk
• Main memory and registers are only storage CPU can
access directly
• Memory unit only sees a stream of:
• addresses + read requests, or
• address + data and write requests
• Protection of memory required to ensure correct
operation
Credit: Wikimedia
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Memory Management Terminilogies
•
Segment: A chunk of memory
assigned to a process.
•
Virtual Address: an address relative
to the start of a process's address
space.
OS
400
0
A
500
0
300
900
1100
C
0
1400
Physical Address
memory
0
Virtual Address
• Physical Address: a real address in
Memory
2000
B
2400
400
Segments
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Where does Address Come From
•
How do programs generate instruction and data
addresses?
•
Compile time: The compiler generates the exact
physical location in memory starting from some xed
starting position k. The OS does nothing.
•
Load time: Compiler generates an address, but at
load time the OS determines the process' starting
position. Once the process loads, it does not move
in memory.
•
Execution time: Compiler generates an address,
and OS can place it any where it wants in memory.
Multistep processing of a user program
Operativsystem: Memory Management
fi
Mohammad Rajiullah
kau.se/cs
Logical Vs. Physical Address
• Logical address
address
generated by the CPU; also referred to as virtual
• Physical address address seen by the memory unit
• Logical and physical addresses are the same in compile-time and loadtime address-binding schemes; logical (virtual) and physical addresses
di er in execution-time address-binding scheme
ff
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
•
•
•
•
•
•
•
Uniprogramming
OS gets a xed part of memory (highest memory in DOS).
One process executes at a time.
Process is always loaded starting at address 0.
Process executes in a contiguous section of memory.
Compiler can generate physical addresses.
Maximum address = Memory Size - OS Size
OS is protected from process by checking addresses used by process.
Operativsystem: Memory Management
fi
Mohammad Rajiullah
kau.se/cs
Uniprogramming
Memory
0
Memory
0
Memory
0
B
A
C
OS
2200
2400
OS
2200
2400
OS
2200
2400
Process A, B and C
Simple, but does not allow for overlap of I/O and computation.
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Multiprogramming
• Transparency:
• We want multiple processes to coexist in memory.
• No process should be aware that memory is shared.
• Processes should not care what physical portion of memory they are assigned to.
• Safety:
• Processes must not be able to corrupt each other.
• Processes must not be able to corrupt the OS.
• E ciency:
• Performance of CPU and memory should not be degraded badly due to sharing.
ffi
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
•
Assume at compile/link time that the
process starts at 0 with a maximum
address = memory size - OS size.
Load a process by allocating a
contiguous segment of memory in
which the process ts.
Memory
Mohammad Rajiullah
0
Memory
400
C
Operativsystem: Memory Management
400
A
A
900
OS
0
B
B
The rst (smallest) physical address
of the process is the base address
and the largest physical address the
process can access is the limit
address.
fi
•
Put the OS in the highest memory.
fi
•
•
Relocaltion
2000
2400
C
OS
900
1200
2000
2400
kau.se/cs
Static Relocation:
•
•
•
at load time, the OS adjusts the addresses in a process to re ect its position in memory.
Once a process is assigned a place in memory and starts executing it, the OS cannot move it.
Dynamic Relocation:
•
•
•
hardware adds relocation register (base) to virtual address to get a physical address;
hardware compares address with limit register (address must be less than limit).
If test fails, the processor takes an address trap and ignores the physical address.
Mohammad Rajiullah
kau.se/cs
Operativsystem: Memory Management
fl
•
Relocation
Dynamic Relocation
Advantages:
•
•
•
•
OS can easily move a process during execution.
OS can allow a process to grow over time.
Simple, fast hardware: two special registers, an add, and a compare.
Disadvantages:
•
•
•
•
•
Slows down hardware due to the add on every memory reference.
Can't share memory (such as program text) between processes.
Process is still limited to physical memory size.
Degree of multiprogramming is very limited since all memory of all active processes must t in memory.
Complicates memory management.
Mohammad Rajiullah
kau.se/cs
Operativsystem: Memory Management
fi
•
Relocation: Properties
• Transparency: processes are largely unaware of sharing.
• Safety: each memory reference is checked.
• E ciency: memory checks and virtual to physical address translation are fast as they
are done in hardware, BUT if a process grows, it may have to be moved which is very
slow.
ffi
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Recap
Uniprogramming
Memory
0
Static Relocation
OS
Contiguous allocation
400
0
A
500
0
300
900
1100
C
0
1400
Physical Address
Dynamic Relocation
Virtual Address
•
•
•
•
2000
B
2400
400
Segments
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Dynamic Loading
•
•
•
•
•
•
The entire program does not need to be in memory to execute
Routine is not loaded until it is called
Better memory-space utilization; unused routine is never loaded
All routines kept on disk in relocatable load format
Useful when large amounts of code are needed to handle infrequently occurring cases
No special support from the operating system is required
• Implemented through program design
• OS can help by providing libraries to implement dynamic loading
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
• Static linking
Dynamic Linking
system libraries and program code combined by the
loader into the binary program image
• Dynamic linking linking postponed until execution time
• Dynamic linking is particularly useful for libraries
• System also known as shared libraries
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Memory Allocation
• As processes enter the system, grow, and terminate, the OS must keep
track of which memory is available and utilized.
• Holes: pieces of free memory (shaded above in gure)
• Given a new process, the OS must decide which hole to use for
the process
Operativsystem: Memory Management
fi
Mohammad Rajiullah
kau.se/cs
Memory Allocation Policies
•
First-Fit: allocate the rst one in the list in which the process ts. The
search can start with the rst hole, or where the previous rst- t search
ended.
•
Best-Fit: Allocate the smallest hole that is big enough to hold the process.
The OS must search the entire list or store the list sorted by size hole list.
•
Worst-Fit: Allocate the largest hole to the process. Again the OS must
search the entire list or keep the list sorted.
•
Simulations show rst- t and best- t usually yield better storage
utilization than worst- t; rst- t is generally faster than best- t.
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
fi
Operativsystem: Memory Management
fi
Mohammad Rajiullah
kau.se/cs
Fragmentation
• External Fragmentation
total memory space exists
to satisfy a request, but it is not contiguous
• Internal Fragmentation
allocated memory may be
slightly larger than requested memory; this size
di erence is memory internal to a partition, but not
being used
• Consider a process of size 8846 bytes and a block
of size 8848 bytes
it is more e cient to allocate the process the
entire 8848 block than it is to keep track of 2 free
bytes
• Internal fragmentation exists when memory
internal to a partition that is wasted
Operativsystem: Memory Management
ffi
ff
Mohammad Rajiullah
kau.se/cs
Compaction
• Reduce external fragmentation by compaction
• Shu e memory contents to place all free memory together in one large
block
• Compaction is possible only if relocation is dynamic, and is done at
execution time
OS
OS
20K hole
Process 1
Process 1
Process 2
10K hole
Process 2
Free 10K
ffl
Mohammad Rajiullah
Free
20+10+10=
40K
Operativsystem: Memory Management
kau.se/cs
Swapping
• Roll out a process to disk, releasing all the memory it holds.
• When process becomes active again, the OS must reload it in memory.
•
With static relocation, the process must be put in the same position.
• With dynamic relocation, the OS nds a new position in memory for the
process and updates the relocation and limit registers.
• If swapping is part of the system, compaction is easy to add.
Operativsystem: Memory Management
fi
Mohammad Rajiullah
kau.se/cs
Another possible solution is non-contiguos memory allocation
Paging: motivation & features
90/10 rule: Processes spend 90% of their time accessing 10% of their space in memory.
=> Keep only those parts of a process in memory that are actually being used
• Pages greatly simplify the hole tting problem
• The logical memory of the process is contiguous, but pages need
not be allocated contiguously in memory.
• By dividing memory into xed size pages, we can eliminate external fragmentation.
• Paging does not eliminate internal fragmentation
fi
Operativsystem: Memory Management
fi
Mohammad Rajiullah
kau.se/cs
Paging
Processes typically do not use their entire space in memory all the time.
Paging
1. divides and assigns processes to xed sized pages,
2. then selectively allocates pages to frames in memory, and
3. manages (moves, removes, reallocates) pages in memory.
Operativsystem: Memory Management
fi
Mohammad Rajiullah
kau.se/cs
Paging: Example
Mapping pages in logical memory to frames in physical memory
Logical memory
Frames in memory
A0
F0
OS
A1
F1
OS
A2
F2
A0
A3
F3
A5
A4
F4
A4
A5
F5
Process A in 6 pages
F6
A1
F7
0
400
800
1200
1600
F8
A3
2000
F11 A2
2400
F9
f10
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Paging Hardware
•
Problem: How do we nd addresses when pages are not allocated contiguously in memory?
•
Virtual Address:
• Processes use a virtual (logical) address to name memory locations.
• Process generates contiguous, virtual addresses from 0 to size of the process.
• The OS lays the process down on pages and the paging hardware translates virtual
addresses to actual physical addresses in memory.
• In paging, the virtual address identi es the page and the page o set.
• page table keeps track of the page frame in memory in which the page is
located.
ff
fi
Operativsystem: Memory Management
fi
Mohammad Rajiullah
kau.se/cs
Paging Hardware
'
Translating a virtual address to a physical address.
Physical address
Logical address
CPU
f
p d
d
frame e
0x000
d
frame ff
0x f
ff
frame g
Page table
physical memory
ff
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Paging Hardware
• Paging is a form of dynamic relocation, where each virtual address is bound by the
paging hardware to a physical address.
• Think of the page table as a set of relocation registers, one for each frame.
• Mapping is invisible to the process; the OS maintains the mapping and the hardware
does the translation.
• Protection is provided with the same mechanisms as used in dynamic relocation.
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Address Translation Scheme
Address generated by CPU is divided into:
•
Page number (p) used as an index into a page table which contains base address of each
page in physical memory
•
Page o set (d) combined with base address to de ne the physical memory address that is
sent to the memory unit
page number
p
d
m -n
n
For given logical address space 2m and page size 2n
Operativsystem: Memory Management
fi
Mohammad Rajiullah
ff
•
page offset
kau.se/cs
Paging model of logical and physical memory
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Paging Example
Logical address: n = 2 and m = 4. Using a page size of 4 bytes and a
physical memory of 32 bytes (8 frames)
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Paging Example
frame
t
Page
⑧
O
page
4
k
O
⑨
I
m= 4
r
m
E
16
i
↑
=
5
3
2
~
page
pagel
N
24
logical
byte
memory22
byte page
&
P
->
4
first
page
E
e
u
↓
logical
-
2
2
2
2
bis
t
P
page
france
2
fratal
last I bits
address
X
Mohammad Rajiullah
i
2
->
13
Of
-
d
freet
t
2
u
frame
Y Ist item
3
Operativsystem: Memory Management
kau.se/cs
Paging example
Consider a machine with 64 MB physical memory and a 32-bit virtual address space. If
the page size is 4KB, what is the approximate size of the page table?
210 = 1 K
220 = 1 M
230 = 1 G
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
•
Paging Example (more)
Frames in Memory
How big is the page table?
• 16 entries (256 memory bytes / 16 byte pages)
•
How many bits for an address, assuming we can
address 1 byte increments?
• 8 bits ( to address 256 bytes)
•
What is p, and d?
Page table
A0
0
2
A1
1
6
A2
2
A3
3
11
9
f0
f1
Page size = 16 bytes
32 bytes
A0 f2
f3
64 bytes
f4
F5
Memory size = 256 bytes
• 4 bits for page and 4 for o set
96 bytes
A1 F6
F7
128 bytes
F8
Given virtual address 24, do the virtual to
physical translation
A3 F9
• Page p=1, o set d=8
A2 F11
160 bytes
F10
192 bytes
F12
• Frame f=6, o ser d=8
F13
224 bytes
F14
Operativsystem: Memory Management
ff
ff
Mohammad Rajiullah
ff
•
Virtual memory
0
F15
kau.se/cs
256 bytes
•
Paging Example (more)
Frames in Memory
How big is the page table?
• 16 entries (256 memory bytes / 16 byte pages)
•
How many bits for an address, assuming we can
address 1 byte increments?
• 8 bits ( to address 256 bytes)
•
0
2
A1
1
6
A2
2
A3
3
11
9
f0
f1
f3
64 bytes
f4
F5
Page size = 16 bytes
32 bytes
A0 f2
96 bytes
A1 F6
F7
128 bytes
F8
Given virtual address 24, do the virtual to
physical translation
A3 F9
• Page p=1, o set d=8
A2 F11
160 bytes
F10
192 bytes
F12
• Frame f=6, o ser d=8
F13
224 bytes
F14
Operativsystem: Memory Management
ff
Mohammad Rajiullah
ff
•
Page table
A0
Memory size = 256 bytes
4 bits for page and 4 for o set
ff
•
What is p, and d?
Virtual memory
0
F15
kau.se/cs
256 bytes
Virtual address to physical address
Given
•
•
•
•
•
•
•
1
2
Virtual address = 4GB
Physical address = 64 MB
Page size= 4 KB
No of pages=?
No of frames=?
No of entries in page table=?
Size of page table=?
Mohammad Rajiullah
Operativsystem: Memory Management
=2
220 = 1 M
22 = 4
230 = 1 G
3
2 =8
240 = 1 T
24 = 16
5
2 = 32
26 = 64
7
2 = 128
28 = 256
9
2 = 512
210 = 1024 or 1K
kau.se/cs
Making Paging Efficient
• How should we store the page table?
• Registers: Advantages? Disadvantages?
• Memory: Advantages? Disadvantages?
• Translation look-aside bu er (TLB): a fast fully associative memory that
stores page numbers (key) and the frame (value) in which they are
stored.
• if memory accesses have locality, address translation has locality too.
• typical TLB sizes range from 8 to 2048 entries.
Operativsystem: Memory Management
ff
Mohammad Rajiullah
kau.se/cs
Paging Hardware with TLB
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Effective Access Time
• Hit ratio percentage of times that a page number is found in the TLB
• An 80% hit ratio means that we nd the desired page number in the TLB 80% of the time.
• Suppose that 10 nanoseconds to access memory.
• If we nd the desired page in TLB then a mapped-memory access take 10 ns
• Otherwise we need two memory access so it is 20 ns
•E
ective Access Time (EAT)
• EAT = 0.80 x 10 + 0.20 x 20 = 12 nanoseconds
• implying 20% slowdown in access time
• Consider
amore realistic hit ratio of 99%,
• EAT = 0.99 x 10 + 0.01 x 20 = 10.1ns
• implying only 1% slowdown in access time.
Operativsystem: Memory Management
fi
fi
ff
Mohammad Rajiullah
kau.se/cs
Memory Protection
v: valid bit that says the entry is up-to-date
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Sharing
•
Paging allows sharing of memory across processes since memory used by a process no longer
needs to be contiguous.
•
Shared code must be reentrant, which means the processes that are using it cannot change it
(e.g., no data in reentrant code).
•
•
Sharing of pages is similar to the way threads share text and memory with each other.
•
The user program (e.g., emacs) marks text segment of a program as reentrant with a system
call.
•
The OS keeps track of available reentrant code in memory and reuses them if a new process
requests the same program.
•
Can greatly reduce overall memory requirements for commonly used applications.
A shared page may exist in di erent parts of the virtual address space of each process, but the
virtual addresses map to the same physical address.
Operativsystem: Memory Management
ff
Mohammad Rajiullah
kau.se/cs
Sharing Example
Sharing of standard C library in a paging environment.
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Structure of the Page Table
•
•
•
•
•
•
•
Memory structures for paging can get huge using straight-forward methods
Consider a 32-bit logical address space as on modern computers
Page size of 4 KB (212)
Page table would have 1 million entries (232 / 212)
If each entry is 4 bytes  each process 4 MB of physical address space for the page table alone
Don’t want to allocate that contiguously in main memory
One simple solution is to divide the page table into smaller units
•
•
•
Hierarchical Paging
Hashed Page Tables
Inverted Page Tables
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Hierarchical Page Table
• Break up the logical address space
into multiple page tables
• A simple technique is a two-level
page table
• We then page the page table
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Swapping
•
•
Moving processes between primary memory and disk
High context-switch time
Standard
swapping of
two processes
using a disk
as a backing
store.
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Swapping with Paging
Mohammad Rajiullah
Operativsystem: Memory Management
kau.se/cs
Paging Summary
•
Paging is a big improvement over relocation:
• They eliminate the problem of external fragmentation and therefore the
need for compaction.
• They allow sharing of code pages among processes, reducing overall memory requirements.
• They enable processes to run when they are only partially loaded in main memory.
•
However, paging has its costs:
• Translating from a virtual address to a physical address is more timeconsuming.
• Paging requires hardware support in the form of a TLB to be e cient enough.
• Paging requires more complex OS to maintain the page table.
Operativsystem: Memory Management
ffi
Mohammad Rajiullah
kau.se/cs