Memory Organization IV – Memory Management Unit

Shobhanjana Kalita,
Dept. of CSE, Tezpur University

Slides courtesy: Computer Architecture and Organization, 9th Ed, W. Stallings

Memory Management Unit
 When the processor works on only one program at a time, it is called a
uniprogramming system.
 Main memory divided into two parts: 1. for OS and 2. for program currently being
executed
 In a multiprogramming system, the processor is kept as busy as possible by
allow it to work on multiple programs at a time.
 Main memory divided into two parts: 1. for OS and 2. for all user programs
 “user” part of memory subdivided to accommodate multiple processes.
 The task of subdivision of memory to accommodate multiple programs is carried
out dynamically by the OS and is known as memory management
Swapping
 User may have many programs to run – MMU tries to fit in these in the main
memory
 All of the programs may not be accommodated in the memory
 Programs that are not accommodated, wait for their turn in a queue
 When one program ends, the memory area it holds is released and replaced by
another program in the queue.
 When a program is waiting for an I/O operation from the user – it is
occupying memory space without being executed
 Such a program may be temporarily moved out of the memory, to accommodate
some other program in the queue
 This is called swapping
Logical address
 During swapping a program is moved out, and then is moved back in again to
continue processing
 May not be moved back to the same location in memory that it had occupied before
 Thus, the physical addresses for instructions and data for a program keep changing –
not fixed
 Logical addresses are used to solve this problem
 A physical address is an actual location in main memory
 A logical address is expressed as a location relative to the beginning of the
program.
 Instructions in the program are referred to using only logical addresses.
 MMU converts from logical to physical address by adding the current starting
location of the process (base address) to the logical address.
Partitioning
 Fixed-size partitions – entire memory
area is divided into fixed sized partitions
such that each partition can be occupied by
one program
 Two types-
 Equal size partitions
 Unequal size partitions – when a program
comes in MMU allocates the smallest
available partition that can hold it
 Disadv – memory wastage
Partitioning
 Variable sized partitioning - When a program is brought into memory, it is
allocated exactly as much memory as it requires and no more
 Disadv – Over time it leads to a situation in which there are a lot of small
holes in memory that cannot be efficiently utilized – this is fragmentation
 One solution is compaction - From time to time, the OS shifts processes in
memory to place all the free memory together in one block.
 time-consuming procedure, wasteful of processor time.
Partitioning

Effect of variable
sized partitioning
Paging
 Both unequal fixed-size and variable-size partitions are inefficient in the use of
memory
 Paging is the approach where –
 memory is partitioned into relatively small, equal fixed-size chunks – called
frames
 each program is also divided into small fixed-size chunks of some size – called
pages
 A page of program occupies a single frame in memory
 The entire set of pages of a program may or may not occupy continuous set frames
in memory
 OS maintains a page table for each program, that shows the frame location for
each page of the process
Paging – Page table
• In the page table for each
process, frame location for
each page of the process is
given
• Logical address consists
of a page number and a
relative address within the
page (offset)
• Physical address consists
of a frame number and
relative address (offset)
Paging

Pages for a program

is allocated from a
free frames list
maintained by OS
Demand Paging
 In simple paging – when a program is brought into memory, all its pages are
fitted into free frames in the memory
 In demand paging – each page of a program is brought in only when it is
needed, that is, on demand
 Entire program may not physically be in memory during execution – only a few
pages are loaded as demanded by CPU
 Principle of locality of reference is used in choosing which pages are to be
loaded
 If the program references data on a page not in memory, a page fault is triggered
– OS then loads appropriate page from disk onto memory
 When it brings one page in, it must throw another page out –this is page
replacement
Demand Paging–Virtual Memory
 With demand paging, it is not necessary to load an entire
process into main memory, allowing
 More number of programs to be in the memory
 A program can be larger than all of main memory
 The details of loading the appropriate pages of a program
under execution into the real memory is left to the OS
and the hardware
 The programmer/user simply perceives a very large memory
– part of which is allocated from the disk
 The part of disk memory that is used by programmer/user in
lieu of main memory is referred to as virtual memory.
 Virtual memory allows for very effective
multiprogramming
 Relieves the user of the tight constraints of main memory
Demand Paging – TLB
 Ideally page table should be accommodated in the MMU, however the page
tables may be too large; page table is instead kept in the main memory
 To read a word from memory –
 Translate logical address to physical address using page table
 Page number + offset -> frame address + offset
 Every memory reference requires two physical memory accesses
 To fetch the appropriate page table entry
 To fetch the desired data
 To overcome this – use a special cache for page table entries is included in the
MMU – called translation lookaside buffer (TLB)
 Contains a copy of a small portion of the page table
 Consists of page table entries that corresponds to most recently used pages
Demand Paging – TLB with Cache
Demand Paging
 CPU requests for some logical address (page number + offset)
 MMU consults the TLB to see if requested page table entry is present
 If yes, the physical address obtained (frame number + offset)
 If not, the entry is accessed from a page table in memory
 If physical address is generated (consisting of tag + remainder), cache is consulted
to see if the block containing the word is present
 If so, it is returned to the processor
 If not, the word is retrieved from main memory
 If requested address corresponds to a page not in main memory, then retrieve the
entire page from disk – this called a page fault.
 MMU raises an exception (interrupt), and control is transferred to OS to fetch the page from disk
 If there are no free frames in the memory, replacement policies (LRU, FIFO etc) are used to
determine which page will be replaced