TDC 311 -

clippersdogheartedΛογισμικό & κατασκευή λογ/κού

14 Δεκ 2013 (πριν από 3 χρόνια και 7 μήνες)

57 εμφανίσεις

TDC 311

Memory Management

Why Do We Need Memory Management?


to protect one process from another; the
OS from processes; resident libraries from processes.

Sharing code


what if two different processes want to
share a piece of code?

Logical organization


allows for multiple,
independent modules to reside in memory at the same

Virtual memory


processes are bigger than main
memory. Memory management allows a large process to
partially occupy main memory.

Historical Perspective

Bare Machine

o memory management. User has
entire machine.

Resident Monitor


memory divided into 2 sections

monitor and user. Protection hardware consists of one
fence register.

User memory does not start at location 0. OS starts at 0.
User starts at location n. Are programs compiled/assembled at
location n? Probably not, so need relocation.

Static relocation

IBM 370 style.

Dynamic relocation

CDC 6600 style with relocation register.

Logical address space : What the user sees : 0 to max.

Physical address space : What the OS uses: R+0 to

fence register value R.

Historical Perspective Continued


memory still divided into monitor space and user
space, but you may have more than one job/process, so swap
one in, execute it, swap it out, swap next one, etc.

Overlapped swapping


memory divided into monitor,
buffer1, buffer2, and user space. Keep two jobs in buffer
spaces and swap them in and out of user space as needed.

Multiple partitions


memory is divided into regions, or
. For a job to execute, it is loaded into a free
partition. When done, the job is removed and the partition is
marked free.

Can have fixed (static) sized partitions (MFT) and variable (dynamic)
sized partitions (MVT) (IBM OS/360).


Multiprogramming with Fixed
Sized Tasks


Asking for more memory

Internal fragmentation


Multiprogramming with Variable
Sized Tasks

No fixed sized partitions. Fit job wherever it will fit. Best
fit? First fit? Worst fit?



Garbage collection (compaction).

Simple Paging

Divide memory into fixed sized frames

Divide user process into fixed sized pages

Take a page of user program and plug into empty frame

Each job has its own page table

No external fragmentation, only internal fragmentation

On average, ½ of a frame will be wasted

Shared pages

possible to share pages (like editor,
compiler, database code) if code is reentrant (does not
modify itself).


can add read only / read write / execute bits
to each page in the page table.

Simple Segmentation

Like paging, but user job not broken into fixed sized pages
(physical division) but left as procedures / functions / data


can now place access bits on entire segment.
A code segment can be accessed by many users. A data
segment can be accessed by only one user. Put an array
in a segment by itself and the hardware will check the
array bounds automatically!


same as with paging

Virtual Memory

Why load the entire program into memory at the start?
Just load the necessary routines / modules. Why? Error
routines are seldom called. Don’t always need entire
array, list or table. Certain sections of code may be
rarely used.

Resident set


portion of user process that is currently
in main memory.

Advantages of virtual memory

may have more
processes in memory since only part of each process is
resident. It is possible for a process to be larger than
main memory.

Virtual Memory Continued


if part of a process is not in memory,
process is moved to blocked state, disk I/O reads in
necessary piece of code, process is moved to Ready state,
process resumes execution


removing a piece to satisfy some other
process, but then needing that piece next.

Demand Paging

Page table for each process must also include 1 bit for
each page

this bit tells if page is in memory, or resident

Also, one Modify bit for each page

tells if the page has
been modified

What if page is not resident?

Take page number, look in page table.

Is the page resident? If yes, get frame number. If no,
page fault

Find a free frame. No free frame? Call replacement algorithm.

Load desired page from disk into memory.

When disk read completed, update page table.

Restart instruction.

Paging Issues

Page size

the smaller the page, the less internal
fragmentation, but the bigger the page table.

If a page is 512 bytes in size (2
) and we are using a
memory space 2
, we will need 2

page table entries per
process! How do you handle such a large page table?
Keep only part of the page table resident, or use an
inverted page table.

Typical page sizes: IBM 370: 2048 / 4096; VAX: 512; INTEL
486: 4096; Motorola 68040: 4096.

The paging process can be time intensive. Make sure the
paging process is fast and doesn’t occur constantly.

Pure demand paging vs.

Page Replacement Algorithms

If you need to bring a page in, but there are no available
frames, you have to remove a page and put it back to disk.
Who do you remove?

Let’s say we are given the following page reference string:

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

and 3 frames of memory

FIFO (first in first out):

LRU (least recently used):

Optimal (theoretical):

Other Issues

How many pages at one time do you give to each process
resident set size
)? Can the resident set size change as
the process’s demands grow and shrink?

When you take a page out to make room for another
page, check its Modify bit. If it has not been set, the page
has not been modified (is not
) and there is no need
to write the page back to disk.

There is also demand segmentation.


Windows NT

Page size = 4096 bytes. Segmentation and paging are
optional and in combinations.

With unsegmented memory, user’s virtual space is 2

= 4

With segmentation, there is a 16
bit segment reference
(of which 2 bits are used for protection) and a 32 bit
offset, for 2

= 64Tb !

Paging is done via 2 levels

first level is a page directory
containing 1024 entries. Each entry is a page table with
1024 entries. Each entry is 4 Kb.



Linux uses either 3
level paging or the more recent 4
level paging

From IBM’s Explore the Linux Memory Model (3



In 64
bit processors:

21 MSBs are unused

13 LSBs are represented by page offset

The remaining 30 bits are divided into

10 bits for Page Table

10 bits for Page Global Directory

10 bits for Page Middle Directory

Thus, 43 bits are used for addressing. This yields 2

bytes of available virtual memory space