CS 241 Section Week #9

(04/09/09)
 Topics
• LMP2 Overview
• Memory Management
• Virtual Memory
• Page Tables
LMP2 Overview
 LMP2 Overview
• LMP2 attempts to encode or decode a
number of files the following way:
– encode: %> ./mmap -e -b16 file1 [file2 ...]
– decode: %> ./mmap -d -b8 file1 [file2 ...]
• It has the following parameters:
– It reads whether it has to encode (‘-e’) or
decode(‘-d’);
– the number of bytes (rw_units) for each
read/write from the file;
 LMP1 Overview
• You have TWO weeks to complete and submit
LMP2. We have divided LMP2 into two
stages:
– Stage 1:
• Implement a simple virtual memory.
• It is recommended you implement the my_mmap()
function during this week.
• You will need to complete various data structures to
deal with the file mapping table, the page table, the
physical memory, etc.
 LMP1 Overview
• You have TWO weeks to complete and submit
LMP2. We have divided LMP2 into two stages:
– Stage 2
• Implement various functions for memory mapped files
including:
– my_mread() , my_mwrite() and my_munmap()
• Handle page faults in your my_mread() and
my_mwrite() functions
• Implement two simple manipulations on files:
– encoding
– decoding
Memory Management
 Memory
• Contiguous allocation and compaction

• Paging and page replacement algorithms

 Fragmentation
• External Fragmentation
– Free space becomes divided into many small pieces
– Caused over time by allocating and freeing the storage of
different sizes

• Internal Fragmentation
– Result of reserving space without ever using its part
– Caused by allocating fixed size of storage
 Contiguous Allocation
• Memory is allocated in monolithic segments
or blocks

• Public enemy #1: external fragmentation

– We can solve this by periodically rearranging the
contents of memory
 Storage Placement Algorithms
• Best Fit
– Produces the smallest leftover hole
– Creates small holes that cannot be used
 Storage Placement Algorithms
• Best Fit
– Produces the smallest leftover hole
– Creates small holes that cannot be used
• First Fit
– Creates average size holes
 Storage Placement Algorithms
• Best Fit
– Produces the smallest leftover hole
– Creates small holes that cannot be used
• First Fit
– Creates average size holes
• Worst Fit
– Produces the largest leftover hole
– Difficult to run large programs
 Storage Placement Algorithms
• Best Fit
– Produces the smallest leftover hole
– Creates small holes that cannot be used
• First Fit
– Creates average size holes
• Worst Fit
– Produces the largest leftover hole
– Difficult to run large programs
First-Fit and Best-Fit are better than Worst-Fit in
terms of SPEED and STORAGE UTILIZATION
 Exercise
• Consider a swapping system in which memory
consists of the following hole sizes in memory order:
10KB, 4KB, 20KB, 18KB, 7KB, 9KB, 12KB, and 15KB.
Which hole is taken for successive segment requests
of (a) 12KB, (b) 10KB, (c) 9KB for
– First Fit?
 Exercise
• Consider a swapping system in which memory
consists of the following hole sizes in memory order:
10KB, 4KB, 20KB, 18KB, 7KB, 9KB, 12KB, and 15KB.
Which hole is taken for successive segment requests
of (a) 12KB, (b) 10KB, (c) 9KB for
– First Fit? 20KB, 10KB and 18KB
 Exercise
• Consider a swapping system in which memory
consists of the following hole sizes in memory order:
10KB, 4KB, 20KB, 18KB, 7KB, 9KB, 12KB, and 15KB.
Which hole is taken for successive segment requests
of (a) 12KB, (b) 10KB, (c) 9KB for
– First Fit? 20KB, 10KB and 18KB
– Best Fit?
 Exercise
• Consider a swapping system in which memory
consists of the following hole sizes in memory order:
10KB, 4KB, 20KB, 18KB, 7KB, 9KB, 12KB, and 15KB.
Which hole is taken for successive segment requests
of (a) 12KB, (b) 10KB, (c) 9KB for
– First Fit? 20KB, 10KB and 18KB
– Best Fit? 12KB, 10KB and 9KB
 Exercise
• Consider a swapping system in which memory
consists of the following hole sizes in memory order:
10KB, 4KB, 20KB, 18KB, 7KB, 9KB, 12KB, and 15KB.
Which hole is taken for successive segment requests
of (a) 12KB, (b) 10KB, (c) 9KB for
– First Fit? 20KB, 10KB and 18KB
– Best Fit? 12KB, 10KB and 9KB
– Worst Fit?
 Exercise
• Consider a swapping system in which memory
consists of the following hole sizes in memory order:
10KB, 4KB, 20KB, 18KB, 7KB, 9KB, 12KB, and 15KB.
Which hole is taken for successive segment requests
of (a) 12KB, (b) 10KB, (c) 9KB for
– First Fit? 20KB, 10KB and 18KB
– Best Fit? 12KB, 10KB and 9KB
– Worst Fit? 20KB, 18KB and 15KB
 malloc Revisited
• Free storage is kept as a list of free blocks
– Each block contains a size, a pointer to the next block, and the space
itself
 malloc Revisited
• Free storage is kept as a list of free blocks
– Each block contains a size, a pointer to the next block, and the space
itself

• When a request for space is made, the free list is scanned

until a big-enough block can be found
– Which storage placement algorithm is used?
 malloc Revisited
• Free storage is kept as a list of free blocks
– Each block contains a size, a pointer to the next block, and the space
itself

• When a request for space is made, the free list is scanned

until a big-enough block can be found
– Which storage placement algorithm is used?

• If the block is found, return it and adjust the free list.

Otherwise, another large chunk is obtained from the OS and
linked into the free list
 malloc Revisited (continued)
typedef long Align; /* for alignment to long */

union header { /* block header */

struct {
union header *ptr; /* next block if on free list */
unsigned size; /* size of this block */
} s;
Align x; /* force alignment of blocks */
};

typedef union header Header;

points to next free block

size
 Compaction
• After numerous malloc() and free() calls,
our memory will have many holes
– Total free memory is much greater than that of any
contiguous chunk

• We can compact our allocated memory

– Shift all allocations to one end of memory, and all holes
to the other end

• Temporarily eliminates of external fragmentation

 Compaction (example)

• Lucky that A fit in there! To be sure that there is enough

space, we may want to compact at (d), (e), or (f)
• Unfortunately, compaction is problematic
– It is very costly. How much, exactly?
– How else can we eliminate external fragmentation?
 Paging
• Divide memory into pages of equal size
– We don’t need to assign contiguous chunks

– Internal fragmentation can only occur on the last page

assigned to a process

– External fragmentation cannot occur at all

– Need to map contiguous logical memory addresses to

disjoint pages
 Page Replacement
• We may not have enough space in physical
memory for all pages of every process at the
same time.

• But which pages shall we keep?

– Use the history of page accesses to decide
– Also useful to know the dirty pages
 Page Replacement Strategies
• It takes two disk operations to replace a dirty page,
so:
– Keep track of dirty bits, attempt to replace clean pages first
– Write dirty pages to disk during idle disk time

• We try to approximate the optimal strategy but can

seldom achieve it, because we don’t know what
order a process will use its pages.
– Best we can do is run a program multiple times, and track
which pages it accesses
 Page Replacement Algorithms
• Optimal: last page to be used in the future is removed first

• FIFO: First in First Out

– Based on time the page has spent in main memory

• LRU: Least Recently Used

– Locality of reference principle again

• MRU: most recently used = removed first

– When would this be useful?

• LFU: Least Frequently Used

– Replace the page that is used least often
 Example
• Physical memory size: 4 pages
• Pages are loaded on demand
• Access history: 0 1 2 3 4 0 1 2 3 4 …
– Which algorithm does best here?

• Access history: 0 1 2 3 4 4 3 2 1 0 …
– And here?
Virtual Memory
 Why Virtual Memory?
• Use main memory as a Cache for the Disk
– Address space of a process can exceed physical memory size
– Sum of address spaces of multiple processes can exceed physical
memory
 Why Virtual Memory?
• Use main memory as a Cache for the Disk
– Address space of a process can exceed physical memory size
– Sum of address spaces of multiple processes can exceed physical
memory
• Simplify Memory Management
– Multiple processes resident in main memory.
• Each process with its own address space
– Only “active” code and data is actually in memory
 Why Virtual Memory?
• Use main memory as a Cache for the Disk
– Address space of a process can exceed physical memory size
– Sum of address spaces of multiple processes can exceed physical
memory
• Simplify Memory Management
– Multiple processes resident in main memory.
• Each process with its own address space
– Only “active” code and data is actually in memory
• Provide Protection
– One process can’t interfere with another.
• because they operate in different address spaces.
– User process cannot access privileged information
• different sections of address spaces have different permissions.
 Principle of Locality
• Program and data references within a process
tend to cluster
 Principle of Locality
• Program and data references within a process
tend to cluster
• Only a few pieces of a process will be needed
over a short period of time (active data or
code)
 Principle of Locality
• Program and data references within a process
tend to cluster
• Only a few pieces of a process will be needed
over a short period of time (active data or
code)
• Possible to make intelligent guesses about
which pieces will be needed in the future
 Principle of Locality
• Program and data references within a process
tend to cluster
• Only a few pieces of a process will be needed
over a short period of time (active data or
code)
• Possible to make intelligent guesses about
which pieces will be needed in the future
• This suggests that virtual memory may work
efficiently
 VM Address Translation
• Parameters
– P = 2p = page size (bytes).
– N = 2n = Virtual address limit
– M = 2m = Physical address limit
n–1 p p–1 0
virtual page number page offset virtual address

address translation

m–1 p p–1 0
physical page number page offset physical address

Page offset bits don’t change as a result of translation

Page Table

• Keeps track of what pages

are in memory
Page Table

• Keeps track of what pages

are in memory
• Provides a mapping from
virtual address to physical
address
 Handling a Page Fault
• Page fault
– Look for an empty page in RAM
• May need to write a page to disk and free it
 Handling a Page Fault
• Page fault
– Look for an empty page in RAM
• May need to write a page to disk and free it
– Load the faulted page into that empty page
 Handling a Page Fault
• Page fault
– Look for an empty page in RAM
• May need to write a page to disk and free it
– Load the faulted page into that empty page
– Modify the page table
 Addressing
• 64MB RAM (2^26)
 Addressing
• 64MB RAM (2^26)
• 2^31 (2GB) total memory

Virtual Address (31 bits)

 Addressing
• 64MB RAM (2^26)
• 2^31 (2GB) total memory
• 4KB page size (2^12)

Virtual Address (31 bits)

 Addressing
• 64MB RAM (2^26)
• 2^31 (2GB) total memory
• 4KB page size (2^12)
• So we need 2^12 for the offset, we can use
the remainder bits for the page

Virtual Address (31 bits)

Virtual Page number (19 bits) Page offset (12 bits)

 Addressing
• 64MB RAM (2^26)
• 2^31 (2GB) total memory
• 4KB page size (2^12)
• So we need 2^12 for the offset, we can use
the remainder bits for the page
– 19 bits, we have 2^19 pages (524288 pages)
Virtual Address (31 bits)

Virtual Page number (19 bits) Page offset (12 bits)

 Address Conversion
• That 19bit page address can be optimized in a variety of
ways
– Translation Look-aside Buffer
Translation Lookaside Buffer (TLB)
• Each virtual memory reference can cause two
physical memory accesses
– One to fetch the page table
– One to fetch the data
Translation Lookaside Buffer (TLB)
• Each virtual memory reference can cause two
physical memory accesses
– One to fetch the page table
– One to fetch the data
• To overcome this problem a high-speed cache
is set up for page table entries
Translation Lookaside Buffer (TLB)
• Each virtual memory reference can cause two
physical memory accesses
– One to fetch the page table
– One to fetch the data
• To overcome this problem a high-speed cache
is set up for page table entries
• Contains page table entries that have been
most recently used (a cache for page table)
Translation Lookaside Buffer (TLB)
 Address Conversion
• That 19bit page address can be optimized in a variety of
ways
– Translation Look-aside Buffer
– Multilevel Page Table
 Multilevel Page Tables
• Given:
– 4KB (212) page size
– 32-bit address space
– 4-byte PTE
 Multilevel Page Tables
• Given:
– 4KB (212) page size
– 32-bit address space
– 4-byte PTE
• Problem:
– Would need a 4 MB page table!
• 220 *4 bytes
 Multilevel Page Tables
• Given:
– 4KB (212) page size
– 32-bit address space
– 4-byte PTE
• Problem:
– Would need a 4 MB page table!
• 220 *4 bytes
• Common solution
– multi-level page tables
– e.g., 2-level table (P6)
• Level 1 table: 1024 entries, each
of which points to a Level 2 page
table.
• Level 2 table: 1024 entries, each
of which points to a page
Summary: Multi-level Page Tables
•Instead of one large table, keep a
tree of tables
–Top-level table stores pointers to
lower level page tables

•First n bits of the page number ==

index of the top-level page table

•Second n bits == index of the 2nd-

level page table

•Etc.
 Example: Two-level Page Table
• 32-bit address space (2GB)
 Example: Two-level Page Table
• 32-bit address space (2GB)

• 12-bit page offset (4kB pages)

 Example: Two-level Page Table
• 32-bit address space (2GB)

• 12-bit page offset (4kB pages)

• 20-bit page address

– First 10 bits index the top-level page table
– Second 10 bits index the 2nd-level page table
– 10 bits == 1024 entries * 4 bytes == 4kB == 1 page
 Example: Two-level Page Table
• 32-bit address space (2GB)

• 12-bit page offset (4kB pages)

• 20-bit page address

– First 10 bits index the top-level page table
– Second 10 bits index the 2nd-level page table
– 10 bits == 1024 entries * 4 bytes == 4kB == 1 page

• Need three memory accesses to read a memory

location
Why use multi-level page tables?
• Split one large page table into many page-sized
chunks
– Typically 4 or 8 MB for a 32-bit address space
Why use multi-level page tables?
• Split one large page table into many page-sized
chunks
– Typically 4 or 8 MB for a 32-bit address space

• Advantage: less memory must be reserved for the

page tables
– Can swap out unused or not recently used tables
Why use multi-level page tables?
• Split one large page table into many page-sized
chunks
– Typically 4 or 8 MB for a 32-bit address space

• Advantage: less memory must be reserved for the

page tables
– Can swap out unused or not recently used tables

• Disadvantage: increased access time on TLB miss

– n+1 memory accesses for n-level page tables
 Address Conversion
• That 19bit page address can be optimized in a variety of
ways
– Translation Look-aside Buffer
– Multilevel Page Table
– Inverted Page Table
Inverted Page Table
•“Normal” page table
–Virtual page number == index
–Physical page number == value
Inverted Page Table
•“Normal” page table
–Virtual page number == index
–Physical page number == value

•Inverted page table

–Virtual page number == value
–Physical page number == index
Inverted Page Table
•“Normal” page table
–Virtual page number == index
–Physical page number == value

•Inverted page table

–Virtual page number == value
–Physical page number == index

•Need to scan the table for the right

value to find the index
–More efficient way: use a hash
table
Example
Virtual Address (1010110)

1010 110
 Example
Page Table Virtual Address (1010110)
Index Present Virtual Addr
0 0 1010 110
1 1
2 0
3 1
4 1
5 0
6 1
 Example
Page Table Virtual Address (1010110)
Index Present Virtual Addr
0 0 1010 110
1 1
2 0
3 1
4 1 1010
5 0
6 1
 Example
Page Table Virtual Address (1010110)
Index Present Virtual Addr
0 0 1010 110
1 1
2 0
3 1
4 1 1010
5 0
6 1
 Example
Page Table Virtual Address (1010110)
Index Present Virtual Addr
0 0 1010 110
1 1
Index == 4 (100)
2 0
3 1
4 1 1010
5 0
6 1
 Example
Page Table Virtual Address (1010110)
Index Present Virtual Addr
0 0 1010 110
1 1
Index == 4 (100)
2 0
3 1
4 1 1010
5 0
6 1
100
 Example
Page Table Virtual Address (1010110)
Index Present Virtual Addr
0 0 1010 110
1 1
Index == 4 (100)
2 0
3 1
4 1 1010
5 0
6 1
100 110
 Example
Page Table Virtual Address (1010110)
Index Present Virtual Addr
0 0 1010 110
1 1
Index == 4 (100)
2 0
3 1
4 1 1010
5 0
6 1
100 110

Physical Address (100110)

 Why use inverted page tables?
• One entry for each page of physical memory
– vs. one per page of logical address space
 Why use inverted page tables?
• One entry for each page of physical memory
– vs. one per page of logical address space

• Advantage: less memory needed to store the

page table
– If address space >> physical memory
 Why use inverted page tables?
• One entry for each page of physical memory
– vs. one per page of logical address space

• Advantage: less memory needed to store the

page table
– If address space >> physical memory

• Disadvantage: increased access time on TLB miss

– Use a hash table to limit the search to one – or at most
a few extra memory accesses
 Summary: Address Conversion
• That 19bit page address can be optimized in a variety of
ways
– Translation Look-aside Buffer
• m – memory cycle,  - hit ratio,  - TLB lookup time
• Effective access time (Eat)
– Eat = (m + )(2m + )(1 – ) = 2m +  – m
– Multilevel Page Table
• Similar to indirect pointers in I-nodes
• Split the 19bits into multiple sections
– Inverted Page Table
• Much smaller, but is slower and more difficult to lookup
Summary: Page Tables
•64kB logical address space
•8 pages * 4kB == 32kB RAM

•16-bit virtual address consists of:

–Page number (4 bits)
–Page offset (12 bits)

•Virtual page number – table index

•Physical frame number – value
•Present bit – is page in memory?
 Summary: Virtual Memory
• RAM is expensive (but fast), disk is cheap (but slow)

• Need to find a way to use the cheaper memory

– Store memory that isn’t frequently used on disk
– Swap pages between disk and memory as needed

• Treat main memory as a cache for pages on disk