VIRTUAL MEMORY

SISTEM OPERASI
S-1 ILMU KOMPUTER USU
VIRTUAL MEMORY
Background
Demand Paging
Copy-on-Write
Page Replacement
Allocation of Frames
Thrashing
Memory-Mapped Files
Allocating Kernel Memory
Other Considerations
OBJECTIVES
To describe the benefits of a virtual memory system
To explain the concepts of demand paging, page-replacement
algorithms, and allocation of page frames
To discuss the principle of the working-set model
To examine the relationship between shared memory and
memory-mapped files
To explore how kernel memory is managed
BACKGROUND
Code needs to be in memory to execute, but entire program rarely
used
 Error code, unusual routines, large data structures

Entire program code not needed at same time

Consider ability to execute partially-loaded program
 Program no longer constrained by limits of physical memory
 Each program takes less memory while running -> more programs run at the same
time
 Increased CPU utilization and throughput with no increase in response time or turnaround time
 Less I/O needed to load or swap programs into memory -> each user program runs
faster
BACKGROUND (CONT.)
Virtual memory – separation of user logical memory from physical
memory
 Only part of the program needs to be in memory for execution
 Logical address space can therefore be much larger than physical address space
 Allows address spaces to be shared by several processes
 Allows for more efficient process creation
 More programs running concurrently
 Less I/O needed to load or swap processes
BACKGROUND (CONT.)
Virtual address space – logical view of how process is stored in
memory
 Usually start at address 0, contiguous addresses until end of space
 Meanwhile, physical memory organized in page frames
 MMU must map logical to physical

Virtual memory can be implemented via:

 Demand paging
 Demand segmentation
VIRTUAL MEMORY THAT IS LARGER THAN PHYSICAL MEMORY
VIRTUAL-ADDRESS SPACE
Usually design logical address space for
stack to start at Max logical address and
grow “down” while heap grows “up”
Maximizes address space use
Unused address space between
the two is hole
 No physical memory needed
until heap or stack grows to a
given new page
Enables sparse address spaces with
holes left for growth, dynamically linked
libraries, etc
System libraries shared via mapping into
virtual address space
Shared memory by mapping pages read-
write into virtual address space
Pages can be shared during fork(),
speeding process creation
SHARED LIBRARY USING VIRTUAL MEMORY
DEMAND PAGING
Could bring entire process into memory at load
time
Or bring a page into memory only when it is
needed
 Less I/O needed, no unnecessary I/O
 Less memory needed
 Faster response
 More users

Similar to paging system with swapping

(diagram on right)
Page is needed  reference to it
 invalid reference  abort
 not-in-memory  bring to memory

Lazy swapper – never swaps a page into

memory unless page will be needed
 Swapper that deals with pages is a pager
BASIC CONCEPTS
With swapping, pager guesses which pages will be used before
swapping out again
Instead, pager brings in only those pages into memory
How to determine that set of pages?
 Need new MMU functionality to implement demand paging

If pages needed are already memory resident

 No difference from non demand-paging

If page needed and not memory resident

 Need to detect and load the page into memory from storage
 Without changing program behavior
 Without programmer needing to change code
VALID-INVALID BIT
With each page table entry a valid–invalid bit is associated
(v  in-memory – memory resident, i  not-in-memory)
Initially valid–invalid bit is set to i on all entries
Example of a page table snapshot:

During MMU address translation, if valid–invalid bit in page table

entry is i  page fault
PAGE TABLE WHEN SOME PAGES ARE NOT IN MAIN MEMORY
PAGE FAULT
If there is a reference to a page, first reference to that page will trap
to operating system:
page fault
1.Operating system looks at another table to decide:
 Invalid reference  abort
 Just not in memory

2.Find free frame

3.Swap page into frame via scheduled disk operation
4.Reset tables to indicate page now in memory
Set validation bit = v
5.Restart the instruction that caused the page fault
STEPS IN HANDLING A PAGE FAULT
ASPECTS OF DEMAND PAGING
Extreme case – start process with no pages in memory
 OS sets instruction pointer to first instruction of process, non-memory-resident -> page fault
 And for every other process pages on first access
 Pure demand paging

Actually, a given instruction could access multiple pages -> multiple page
faults
 Consider fetch and decode of instruction which adds 2 numbers from memory and stores
result back to memory
 Pain decreased because of locality of reference

Hardware support needed for demand paging

 Page table with valid / invalid bit
 Secondary memory (swap device with swap space)
 Instruction restart
INSTRUCTION RESTART
Consider an instruction that could access several different locations
 block move

 auto increment/decrement location

 Restart the whole operation?
 What if source and destination overlap?
PERFORMANCE OF DEMAND PAGING
Stages in Demand Paging (worse case)
1.Trap to the operating system
2.Save the user registers and process state
3.Determine that the interrupt was a page fault
4.Check that the page reference was legal and determine the location of the page on the disk
5.Issue a read from the disk to a free frame:
1. Wait in a queue for this device until the read request is serviced
2. Wait for the device seek and/or latency time
3. Begin the transfer of the page to a free frame

6.While waiting, allocate the CPU to some other user

7.Receive an interrupt from the disk I/O subsystem (I/O completed)
8.Save the registers and process state for the other user
9.Determine that the interrupt was from the disk
10.Correct the page table and other tables to show page is now in memory
11.Wait for the CPU to be allocated to this process again
12.Restore the user registers, process state, and new page table, and then resume the interrupted instruction
 PERFORMANCE OF DEMAND PAGING (CONT.)
Three major activities
 Service the interrupt – careful coding means just several hundred instructions needed
 Read the page – lots of time
 Restart the process – again just a small amount of time

Page Fault Rate 0  p  1

 if p = 0 no page faults
 if p = 1, every reference is a fault

Effective Access Time (EAT)

EAT = (1 – p) x memory access
+ p (page fault overhead
+ swap page out
+ swap page in )
DEMAND PAGING EXAMPLE
Memory access time = 200 nanoseconds
Average page-fault service time = 8 milliseconds
EAT = (1 – p) x 200 + p (8 milliseconds)
= (1 – p x 200 + p x 8,000,000
= 200 + p x 7,999,800
If one access out of 1,000 causes a page fault, then
EAT = 8.2 microseconds.
This is a slowdown by a factor of 40!!
If want performance degradation < 10 percent
 220 > 200 + 7,999,800 x p
20 > 7,999,800 x p
 p < .0000025
 < one page fault in every 400,000 memory accesses
DEMAND PAGING OPTIMIZATIONS
Swap space I/O faster than file system I/O even if on the same device
 Swap allocated in larger chunks, less management needed than file system

Copy entire process image to swap space at process load time

 Then page in and out of swap space
 Used in older BSD Unix

Demand page in from program binary on disk, but discard rather than paging out when
freeing frame
 Used in Solaris and current BSD
 Still need to write to swap space
 Pages not associated with a file (like stack and heap) – anonymous memory
 Pages modified in memory but not yet written back to the file system

Mobile systems
 Typically don’t support swapping
 Instead, demand page from file system and reclaim read-only pages (such as code)
COPY-ON-WRITE
Copy-on-Write (COW) allows both parent and child processes to initially share the same
pages in memory
 If either process modifies a shared page, only then is the page copied

COW allows more efficient process creation as only modified pages are copied
In general, free pages are allocated from a pool of zero-fill-on-demand pages
 Pool should always have free frames for fast demand page execution
 Don’t want to have to free a frame as well as other processing on page fault
 Why zero-out a page before allocating it?

vfork() variation on fork() system call has parent suspend and child using copy-
on-write address space of parent
 Designed to have child call exec()
 Very efficient
BEFORE PROCESS 1 MODIFIES PAGE C
AFTER PROCESS 1 MODIFIES PAGE C
 WHAT HAPPENS IF THERE IS NO FREE FRAME?

Used up by process pages

Also in demand from the kernel, I/O buffers, etc
How much to allocate to each?
Page replacement – find some page in memory, but not really in use,
page it out
 Algorithm – terminate? swap out? replace the page?
 Performance – want an algorithm which will result in minimum number of page faults

Same page may be brought into memory several times

PAGE REPLACEMENT
Prevent over-allocation of memory by modifying page-fault
service routine to include page replacement
Use modify (dirty) bit to reduce overhead of page transfers –
only modified pages are written to disk
Page replacement completes separation between logical
memory and physical memory – large virtual memory can be
provided on a smaller physical memory
NEED FOR PAGE REPLACEMENT
 BASIC PAGE REPLACEMENT
1. Find the location of the desired page on disk

2. Find a free frame:

- If there is a free frame, use it
- If there is no free frame, use a page replacement algorithm to
select a victim frame
- Write victim frame to disk if dirty

3. Bring the desired page into the (newly) free frame; update the page
and frame tables

4. Continue the process by restarting the instruction that caused the trap

Note now potentially 2 page transfers for page fault – increasing EAT
PAGE REPLACEMENT
 PAGE AND FRAME REPLACEMENT ALGORITHMS

Frame-allocation algorithm determines

 How many frames to give each process
 Which frames to replace

Page-replacement algorithm
 Want lowest page-fault rate on both first access and re-access

Evaluate algorithm by running it on a particular string of memory

references (reference string) and computing the number of page faults
on that string
 String is just page numbers, not full addresses
 Repeated access to the same page does not cause a page fault
 Results depend on number of frames available

In all our examples, the reference string of referenced page numbers is

7,0,1,2,0,3,0,4,2,3,0,3,0,3,2,1,2,0,1,7,0,1
GRAPH OF PAGE FAULTS VERSUS THE NUMBER OF FRAMES
FIRST-IN-FIRST-OUT (FIFO) ALGORITHM
Reference string: 7,0,1,2,0,3,0,4,2,3,0,3,0,3,2,1,2,0,1,7,0,1
3 frames (3 pages can be in memory at a time per process)

15 page faults

Can vary by reference string: consider 1,2,3,4,1,2,5,1,2,3,4,5

 Adding more frames can cause more page faults!
 Belady’s Anomaly

How to track ages of pages?

 Just use a FIFO queue
FIFO ILLUSTRATING BELADY’S ANOMALY
OPTIMAL ALGORITHM
Replace page that will not be used for longest period of time
 9 is optimal for the example

How do you know this?

 Can’t read the future

Used for measuring how well your algorithm performs

 LEAST RECENTLY USED (LRU) ALGORITHM
Use past knowledge rather than future
Replace page that has not been used in the most amount of time
Associate time of last use with each page

12 faults – better than FIFO but worse than OPT

Generally good algorithm and frequently used
But how to implement?
LRU ALGORITHM (CONT.)
Counter implementation
 Every page entry has a counter; every time page is referenced through this entry, copy
the clock into the counter
 When a page needs to be changed, look at the counters to find smallest value
 Search through table needed

Stack implementation
 Keep a stack of page numbers in a double link form:
 Page referenced:
 move it to the top
 requires 6 pointers to be changed
 But each update more expensive
 No search for replacement

LRU and OPT are cases of stack algorithms that don’t have Belady’s
Anomaly
USE OF A STACK TO RECORD MOST RECENT PAGE REFERENCES
LRU APPROXIMATION ALGORITHMS
LRU needs special hardware and still slow
Reference bit
 With each page associate a bit, initially = 0
 When page is referenced bit set to 1
 Replace any with reference bit = 0 (if one exists)
 We do not know the order, however

Second-chance algorithm
 Generally FIFO, plus hardware-provided reference bit
 Clock replacement
 If page to be replaced has
 Reference bit = 0 -> replace it
 reference bit = 1 then:
 set reference bit 0, leave page in memory
 replace next page, subject to same rules
SECOND-CHANCE (CLOCK) PAGE-REPLACEMENT ALGORITHM
 ENHANCED SECOND-CHANCE ALGORITHM

Improve algorithm by using reference bit and modify bit (if

available) in concert
Take ordered pair (reference, modify)
1.(0, 0) neither recently used not modified – best page to replace
2.(0, 1) not recently used but modified – not quite as good, must write
out before replacement
3.(1, 0) recently used but clean – probably will be used again soon
4.(1, 1) recently used and modified – probably will be used again
soon and need to write out before replacement
When page replacement called for, use the clock scheme but use the
four classes replace page in lowest non-empty class
 Might need to search circular queue several times
COUNTING ALGORITHMS
Keep a counter of the number of references that have been made to
each page
 Not common

Lease Frequently Used (LFU) Algorithm: replaces page with smallest

count

Most Frequently Used (MFU) Algorithm: based on the argument that

the page with the smallest count was probably just brought in and has
yet to be used
PAGE-BUFFERING ALGORITHMS
Keep a pool of free frames, always
 Then frame available when needed, not found at fault time
 Read page into free frame and select victim to evict and add to free pool
 When convenient, evict victim

Possibly, keep list of modified pages

 When backing store otherwise idle, write pages there and set to non-dirty

Possibly, keep free frame contents intact and note what is in them
 If referenced again before reused, no need to load contents again from disk
 Generally useful to reduce penalty if wrong victim frame selected
APPLICATIONS AND PAGE REPLACEMENT

All of these algorithms have OS guessing about future page access

Some applications have better knowledge – i.e. databases
Memory intensive applications can cause double buffering
 OS keeps copy of page in memory as I/O buffer
 Application keeps page in memory for its own work

Operating system can given direct access to the disk, getting out of
the way of the applications
 Raw disk mode

Bypasses buffering, locking, etc

ALLOCATION OF FRAMES
Each process needs minimum number of frames
Example: IBM 370 – 6 pages to handle SS MOVE instruction:
 instruction is 6 bytes, might span 2 pages
 2 pages to handle from
 2 pages to handle to

Maximum of course is total frames in the system

Two major allocation schemes
 fixed allocation
 priority allocation

Many variations
FIXED ALLOCATION
Equal allocation – For example, if there are 100 frames (after
allocating frames for the OS) and 5 processes, give each process 20
frames
 Keep some as free frame buffer pool

Proportional allocation – Allocate according to the size of process

 Dynamic as degree of multiprogramming, process sizes change
m = 64
si = size of process pi s1 = 10
S =  si s2 = 127
m = total number of frames a1 =
10
´ 62 » 4
137
s
ai = allocation for pi = i  m 127
S a2 = ´ 62 » 57
137
PRIORITY ALLOCATION
Use a proportional allocation scheme using priorities rather than
size

If process Pi generates a page fault,

 select for replacement one of its frames
 select for replacement a frame from a process with lower priority number
 GLOBAL VS. LOCAL ALLOCATION
Global replacement – process selects a replacement frame from
the set of all frames; one process can take a frame from another
 But then process execution time can vary greatly
 But greater throughput so more common

Local replacement – each process selects from only its own set of
allocated frames
 More consistent per-process performance
 But possibly underutilized memory
NON-UNIFORM MEMORY ACCESS
So far all memory accessed equally
Many systems are NUMA – speed of access to memory varies
 Consider system boards containing CPUs and memory, interconnected over a system
bus

Optimal performance comes from allocating memory “close to” the

CPU on which the thread is scheduled
 And modifying the scheduler to schedule the thread on the same system board when
possible
 Solved by Solaris by creating lgroups
 Structure to track CPU / Memory low latency groups
 Used my schedule and pager
 When possible schedule all threads of a process and allocate all memory for that process within the lgroup
THRASHING
If a process does not have “enough” pages, the page-fault rate is very
high
 Page fault to get page
 Replace existing frame
 But quickly need replaced frame back
 This leads to:
 Low CPU utilization
 Operating system thinking that it needs to increase the degree of multiprogramming
 Another process added to the system

Thrashing  a process is busy swapping pages in and out

THRASHING (CONT.)
 DEMAND PAGING AND THRASHING
Why does demand paging work?
Locality model
 Process migrates from one locality to another
 Localities may overlap

Why does thrashing occur?

 size of locality > total memory size
 Limit effects by using local or priority page replacement
LOCALITY IN A MEMORY-REFERENCE PATTERN
WORKING-SET MODEL
  working-set window  a fixed number of page references
Example: 10,000 instructions
WSSi (working set of Process Pi) =
total number of pages referenced in the most recent  (varies in time)
 if  too small will not encompass entire locality
 if  too large will encompass several localities
 if  =   will encompass entire program

D =  WSSi  total demand frames

 Approximation of locality

if D > m  Thrashing

Policy if D > m, then suspend or swap out one of the processes

KEEPING TRACK OF THE WORKING SET
Approximate with interval timer + a reference bit
Example:  = 10,000
 Timer interrupts after every 5000 time units
 Keep in memory 2 bits for each page
 Whenever a timer interrupts copy and sets the values of all reference bits to 0
 If one of the bits in memory = 1  page in working set

Why is this not completely accurate?

Improvement = 10 bits and interrupt every 1000 time units
PAGE-FAULT FREQUENCY
More direct approach than WSS
Establish “acceptable” page-fault frequency (PFF) rate and
use local replacement policy
 If actual rate too low, process loses frame
 If actual rate too high, process gains frame
 WORKING SETS AND PAGE FAULT RATES
n Direct relationship between working set of a process and its page-fault
rate
n Working set changes over time
n Peaks and valleys over time
MEMORY-MAPPED FILES
Memory-mapped file I/O allows file I/O to be treated as routine
memory access by mapping a disk block to a page in memory
A file is initially read using demand paging
 A page-sized portion of the file is read from the file system into a physical page
 Subsequent reads/writes to/from the file are treated as ordinary memory accesses

Simplifies and speeds file access by driving file I/O through memory
rather than read() and write() system calls
Also allows several processes to map the same file allowing the pages
in memory to be shared
But when does written data make it to disk?
 Periodically and / or at file close() time
 For example, when the pager scans for dirty pages
 MEMORY-MAPPED FILE TECHNIQUE FOR ALL I/O

Some OSes uses memory mapped files for standard I/O

Process can explicitly request memory mapping a file via mmap() system
call
 Now file mapped into process address space

For standard I/O (open(), read(), write(), close()),

mmap anyway
 But map file into kernel address space
 Process still does read() and write()
 Copies data to and from kernel space and user space
 Uses efficient memory management subsystem
 Avoids needing separate subsystem

COW can be used for read/write non-shared pages

Memory mapped files can be used for shared memory (although again
via separate system calls)
MEMORY MAPPED FILES
SHARED MEMORY VIA MEMORY-MAPPED I/O
SHARED MEMORY IN WINDOWS API
First create a file mapping for file to be mapped
 Then establish a view of the mapped file in process’s virtual address space

Consider producer / consumer

 Producer create shared-memory object using memory mapping features
 Open file via CreateFile(), returning a HANDLE
 Create mapping via CreateFileMapping() creating a named shared-
memory object
 Create view via MapViewOfFile()
ALLOCATING KERNEL MEMORY
Treated differently from user memory
Often allocated from a free-memory pool
 Kernel requests memory for structures of varying sizes
 Some kernel memory needs to be contiguous
 I.e. for device I/O
BUDDY SYSTEM
Allocates memory from fixed-size segment consisting of physically-
contiguous pages
Memory allocated using power-of-2 allocator
 Satisfies requests in units sized as power of 2
 Request rounded up to next highest power of 2
 When smaller allocation needed than is available, current chunk split into two buddies of
next-lower power of 2
 Continue until appropriate sized chunk available

For example, assume 256KB chunk available, kernel requests 21KB

 Split into AL and AR of 128KB each
 One further divided into BL and BR of 64KB
 One further into CL and CR of 32KB each – one used to satisfy request

Advantage – quickly coalesce unused chunks into larger chunk

Disadvantage - fragmentation
BUDDY SYSTEM ALLOCATOR
 SLAB ALLOCATOR
Alternate strategy
Slab is one or more physically contiguous pages
Cache consists of one or more slabs
Single cache for each unique kernel data structure
 Each cache filled with objects – instantiations of the data structure

When cache created, filled with objects marked as free

When structures stored, objects marked as used
If slab is full of used objects, next object allocated from empty slab
 If no empty slabs, new slab allocated

Benefits include no fragmentation, fast memory request satisfaction

SLAB ALLOCATION
 SLAB ALLOCATOR IN LINUX
For example process descriptor is of type struct task_struct
Approx 1.7KB of memory
New task -> allocate new struct from cache
 Will use existing free struct task_struct

Slab can be in three possible states

1.Full – all used
2.Empty – all free
3.Partial – mix of free and used

Upon request, slab allocator

1.Uses free struct in partial slab
2.If none, takes one from empty slab
3.If no empty slab, create new empty
 SLAB ALLOCATOR IN LINUX (CONT.)
Slab started in Solaris, now wide-spread for both kernel mode and
user memory in various OSes
Linux 2.2 had SLAB, now has both SLOB and SLUB allocators
 SLOB for systems with limited memory
 Simple List of Blocks – maintains 3 list objects for small, medium, large objects
 SLUB is performance-optimized SLAB removes per-CPU queues, metadata stored in
page structure
THANK YOU