Transcript csc432-week15x - Rensselaer Polytechnic Institute: Computer

Rensselaer Polytechnic Institute
CSC 432 – Operating Systems
David Goldschmidt, Ph.D.

A noncontiguous memory allocation scheme
avoids the external fragmentation problem
 Slice up physical memory into
fixed-sized blocks called frames
▪ Sizes typically range from 216 and up!
 Slice up logical memory into
fixed-sized blocks called pages
 Allocate pages into frames
▪ Note that frame size equals page size

When a process of size n pages is ready to
run, operating system finds n free frames

The OS keeps
track of pages
via a page table
== in use
== free
process Pi
main memory
 Page tables map logical memory
addresses to physical memory
addresses

Covers a logical address
space of size 2m with
page size 2n
page number page offset
p
d
(m – n)
(n)

Use page table
caching at the
hardware level
to speed address
translation
 Hardware-level
translation look-aside buffer
(TLB)

Segmentation is a memory-management
scheme that corresponds
to logical segments of a
user program
 A segment resides in a
contiguous block of memory
 Segments are numbered
and referred to by a
<segment-number, offset> pair

Logical segments map to physical memory
4
3
3
0
4
2
0
1
2
1

A segment table maps a <segment-number,
offset> pair to a physical address

Each table entry has:
 A segment base containing
the starting physical address
where the segment resides
 A segment limit specifying
the length of the segment

Only part of a process needs to be loaded
into memory for it to start its execution
 Virtual memory further separates logical memory
and physical memory
 Logical (or virtual) address space can be larger
than physical address space
 Allows physical address space to be shared
by several processes
 Enables quicker process creation

Unused pages are stored on disk

Multiple processes can share common
libraries or data by mapping virtual pages to
shared physical
pages
 More efficient
use of physical
memory space

When a page of memory is requested, if it’s
not in physical memory,
load page from disk
 i.e. on demand
 Less I/O required
 Less physical memory
 Faster user response
times (usually)
 More user processes

Virtual memory policies include:
 The fetch policy governs when a page
should be loaded into memory
 The placement policy specifies where a
page is loaded into physical memory
 The replacement policy determines which
existing page in memory should be replaced

Page allocation:
 In a static allocation scheme,
the number of frames per process is fixed
 In a dynamic allocation scheme,
the number of frames per process varies
 In an equal allocation scheme, all processes
have an equal number of frames
 In a proportional allocation scheme, processes
are allocated frames in proportion to size,
priority, etc.

Associate a valid/invalid bit with
each page table entry
frame #
 Initially set all entries to i
 During address translation,
if valid/invalid bit is v, page
is already in memory
 Otherwise, if bit is i,
a page fault occurs
valid/invalid bit
v
v
v
v
i
….
i
i
page table

Page faults are trapped by the OS:
 When an invalid reference occurs in the page table
 When a page is not yet in the page table

Page fault recovery:
 Get free frame from physical memory
 Swap desired page into free frame
 Reset page table entry
 Set validation bit to v
 Restart instruction that caused the page fault

Page faults
are costly!

The page fault rate p is in the range [0.0, 1.0]:
 If p is 0.0, no page faults occur
 If p is 1.0, every page request is a page fault
 Typically p is very low....

The effective memory-access time is
 (1 – p) x physical-memory-access +
p
x
( page-fault-overhead +
swap-page-out + swap-page-in +
restart-overhead )

Given:
 Memory access time is 200 nanoseconds
 Average page-fault service time is 8 milliseconds

The effective memory access time is
 (1 – p) x 200ns + p
8ms =
 200ns – 200ns p + p x 8,000,000ns =
 200ns + 7,999,800 p
x

If a process does not have enough pages, the
page-fault rate is high, leading to thrashing
 Process is busy swapping pages
in and out of memory
 Low CPU utilization
 Operating system might think
it needs to increase the degree
of multiprogramming!
▪ More processes added, further degrading performance

Remember the Principle of Locality

Future memory references in a given
process will likely be local to previous
memory references
 This phenomenon is called
the principle of locality
 A process executes in
a series of phases, spending
a finite amount of time performing
memory references in each phase

Example graph of page faults versus total
number of allocated frames

Operating system should allocate enough
frames for the current locality of a process:
 What happens when too few
frames are allocated?
 What happens when
too many frames are allocated?

Example of a
single process:

Dynamic page fault frequency scheme
 How do we
identify the victim?

Page replacement algorithms include:
 First-in-first-out (FIFO)
 Optimal (OPT)
apply these algorithms to a
page reference stream
 Least recently used (LRU)
 Least frequently used (LFU)
 Page fault frequency scheme (introduced earlier)
 Working set
 The above is a 3-frame memory
 How many page faults occur if we use
a 4-frame memory instead?

Reference string: 1, 2, 3, 4, 1, 2, 5, 1, 2, 3, 4, 5
 How many page faults occur with a 3-frame
memory?
 How many page faults occur with a 4-frame
memory?

Belady’s Anomaly:
 More frames may lead to more page faults!

Replace pages that will not be used for
the longest amount of time
FIFO:
OPT:

Replace pages that have not been used for
the longest amount of time

Similar to LRU, but replace
least frequently used pages
 Requires usage counts
 Initial page replacements are
swapped out quickly because
their usage counts are 1

For each process, maintain a working set of
pages referenced by the process during the
most recent w page references
 How do we choose w?

Let d be the sum of the sizes of the working
sets of all active processes

Let F be the total number of frames
 If d < F, then the operating system can allow
additional processes to enter the system
 If d > F, then the operating system must suspend
one or more active processes
▪ Otherwise thrashing will occur!

Windows XP uses demand paging with
clustering (which loads pages surrounding
the page fault)
 Processes are assigned a working set minimum
and a working set maximum
 The working set minimum is the minimum
number of pages a process is guaranteed to
have in physical memory
 Likewise, a process may be assigned 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
 Increases the amount
of free memory
 Removes pages from
processes that have pages
in excess of their working
set minimum